localhost ring fixes

Pipeline + MLX Ring worked with 2 nodes but failed to initialize with
3 or more nodes. The MLX ring backend requires each node to know its
specific left and right neighbors in the ring, but the previous
implementation provided a single flat host list shared by all nodes.

With 2 nodes, a flat list [host0, host1] accidentally worked because
each node could find its only neighbor. With 3+ nodes, each node needs
a customized view:
- Rank 0: [self, right_neighbor, placeholder]
- Rank 1: [left_neighbor, self, right_neighbor]
- Rank 2: [placeholder, left_neighbor, self]

Changed MlxRingInstance from `hosts: list[Host]` to
`hosts_by_node: dict[NodeId, list[Host]]` with `ephemeral_port: int`.

Added `get_mlx_ring_hosts_by_node()` which generates per-node host
lists where:
- Self position uses 0.0.0.0 for local binding
- Left/right neighbors use actual connection IPs
- Non-neighbors use 198.51.100.1 (RFC 5737 TEST-NET-2 placeholder)

Also added IP prioritization (en0 > en1 > non-Thunderbolt > any) to
prefer stable network interfaces.

.gitignore

@@ -7,8 +7,6 @@ digest.txt
 # nix
 .direnv/
 
-# IDEA (PyCharm)
-.idea
 
 # xcode / macos
 *.xcuserstate
@@ -16,7 +14,6 @@ digest.txt
 *.xcuserdatad/
 **/.DS_Store
 app/EXO/build/
-dist/
 
 
 # rust

README.md

@@ -61,10 +61,10 @@ Devices running exo automatically discover each other, without needing any manua
 
 There are two ways to run exo:
 
-### Run from Source (macOS)
+### Run from Source (Mac & Linux)
 
 **Prerequisites:**
-- [brew](https://github.com/Homebrew/brew) (for simple package management on macOS)
+- [brew](https://github.com/Homebrew/brew) (for simple package management on MacOS)
   
   ```bash
   /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
@@ -98,62 +98,6 @@ uv run exo
 
 This starts the exo dashboard and API at http://localhost:52415/
 
-### Run from Source (Linux)
-
-**Prerequisites:**
-
-- [uv](https://github.com/astral-sh/uv) (for Python dependency management)
-- [node](https://github.com/nodejs/node) (for building the dashboard) - version 18 or higher
-- [rust](https://github.com/rust-lang/rustup) (to build Rust bindings, nightly for now)
-
-**Installation methods:**
-
-**Option 1: Using system package manager (Ubuntu/Debian example):**
-```bash
-# Install Node.js and npm
-sudo apt update
-sudo apt install nodejs npm
-
-# Install uv
-curl -LsSf https://astral.sh/uv/install.sh | sh
-
-# Install Rust (using rustup)
-curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
-rustup toolchain install nightly
-```
-
-**Option 2: Using Homebrew on Linux (if preferred):**
-```bash
-# Install Homebrew on Linux
-/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
-
-# Install dependencies
-brew install uv node
-
-# Install Rust (using rustup)
-curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
-rustup toolchain install nightly
-```
-
-**Note:** The `macmon` package is macOS-only and not required for Linux.
-
-Clone the repo, build the dashboard, and run exo:
-
-```bash
-# Clone exo
-git clone https://github.com/exo-explore/exo
-
-# Build dashboard
-cd exo/dashboard && npm install && npm run build && cd ..
-
-# Run exo
-uv run exo
-```
-
-This starts the exo dashboard and API at http://localhost:52415/
-
-**Important note for Linux users:** Currently, exo runs on CPU on Linux. GPU support for Linux platforms is under development. If you'd like to see support for your specific Linux hardware, please [search for existing feature requests](https://github.com/exo-explore/exo/issues) or create a new one.
-
 ### macOS App
 
 exo ships a macOS app that runs in the background on your Mac.
@@ -166,47 +110,6 @@ Download the latest build here: [EXO-latest.dmg](https://assets.exolabs.net/EXO-
 
 The app will ask for permission to modify system settings and install a new Network profile. Improvements to this are being worked on.
 
-#### Uninstalling the macOS App
-
-The recommended way to uninstall is through the app itself: click the menu bar icon → Advanced → Uninstall. This cleanly removes all system components.
-
-If you've already deleted the app, you can run the standalone uninstaller script:
-
-```bash
-sudo ./app/EXO/uninstall-exo.sh
-```
-
-This removes:
-- Network setup LaunchDaemon
-- Network configuration script
-- Log files
-- The "exo" network location
-
-**Note:** You'll need to manually remove EXO from Login Items in System Settings → General → Login Items.
-
----
-
-### Enabling RDMA on macOS
-
-RDMA is a new capability added to macOS 26.2. It works on any Mac with Thunderbolt 5 (M4 Pro Mac Mini, M4 Max Mac Studio, M4 Max MacBook Pro, M3 Ultra Mac Studio).
-
-Note that on Mac Studio, you cannot use the Thunderbolt 5 port next to the Ethernet port.
-
-To enable RDMA on macOS, follow these steps:
-
-1. Shut down your Mac.
-2. Hold down the power button for 10 seconds until the boot menu appears.
-3. Select "Options" to enter Recovery mode.
-4. When the Recovery UI appears, open the Terminal from the Utilities menu.
-5. In the Terminal, type:
-   ```
-   rdma_ctl enable
-   ```
-   and press Enter.
-6. Reboot your Mac.
-
-After that, RDMA will be enabled in macOS and exo will take care of the rest.
-
 ---
 
 ### Using the API

app/EXO/EXO/ContentView.swift

@@ -17,11 +17,9 @@ struct ContentView: View {
     @State private var deletingInstanceIDs: Set<String> = []
     @State private var showAllNodes = false
     @State private var showAllInstances = false
-    @State private var showAdvanced = false
     @State private var showDebugInfo = false
     @State private var bugReportInFlight = false
     @State private var bugReportMessage: String?
-    @State private var uninstallInProgress = false
 
     var body: some View {
         VStack(alignment: .leading, spacing: 12) {
@@ -51,7 +49,7 @@ struct ContentView: View {
 
     private var topologySection: some View {
         Group {
-            if let topology = stateService.latestSnapshot?.topologyViewModel(localNodeId: stateService.localNodeId), !topology.nodes.isEmpty {
+            if let topology = stateService.latestSnapshot?.topologyViewModel(), !topology.nodes.isEmpty {
                 TopologyMiniView(topology: topology)
             }
         }
@@ -195,7 +193,11 @@ struct ContentView: View {
                 Divider()
                     .padding(.vertical, 4)
             }
-            advancedSection
+            controlButton(title: "Check for Updates") {
+                updater.checkForUpdates()
+            }
+            .padding(.bottom, 8)
+            debugSection
                 .padding(.bottom, 8)
             controlButton(title: "Quit", tint: .secondary) {
                 controller.stop()
@@ -204,33 +206,6 @@ struct ContentView: View {
         }
     }
 
-    private var advancedSection: some View {
-        VStack(alignment: .leading, spacing: 6) {
-            HStack {
-                Text("Advanced")
-                    .font(.caption)
-                    .foregroundColor(.secondary)
-                Spacer()
-                collapseButton(isExpanded: $showAdvanced)
-            }
-            .animation(nil, value: showAdvanced)
-            if showAdvanced {
-                VStack(alignment: .leading, spacing: 2) {
-                    HoverButton(title: "Check for Updates", small: true) {
-                        updater.checkForUpdates()
-                    }
-                    debugSection
-                    HoverButton(title: "Uninstall", tint: .red, small: true) {
-                        showUninstallConfirmationAlert()
-                    }
-                    .disabled(uninstallInProgress)
-                }
-                .transition(.opacity)
-            }
-        }
-        .animation(.easeInOut(duration: 0.25), value: showAdvanced)
-    }
-
     private func controlButton(title: String, tint: Color = .primary, action: @escaping () -> Void) -> some View {
         HoverButton(title: title, tint: tint, trailingSystemImage: nil, action: action)
     }
@@ -353,15 +328,15 @@ struct ContentView: View {
     }
 
     private var debugSection: some View {
-        VStack(alignment: .leading, spacing: 4) {
-            HoverButton(
-                title: "Debug Info",
-                tint: .primary,
-                trailingSystemImage: showDebugInfo ? "chevron.up" : "chevron.down",
-                small: true
-            ) {
-                showDebugInfo.toggle()
+        VStack(alignment: .leading, spacing: 6) {
+            HStack {
+                Text("Debug Info")
+                    .font(.caption)
+                    .foregroundColor(.secondary)
+                Spacer()
+                collapseButton(isExpanded: $showDebugInfo)
             }
+            .animation(nil, value: showDebugInfo)
             if showDebugInfo {
                 VStack(alignment: .leading, spacing: 4) {
                     Text("Version: \(buildTag)")
@@ -377,7 +352,6 @@ struct ContentView: View {
                     sendBugReportButton
                         .padding(.top, 6)
                 }
-                .padding(.leading, 8)
                 .transition(.opacity)
             }
         }
@@ -473,88 +447,6 @@ struct ContentView: View {
         bugReportInFlight = false
     }
 
-    private func showUninstallConfirmationAlert() {
-        let alert = NSAlert()
-        alert.messageText = "Uninstall EXO"
-        alert.informativeText = """
-            This will remove EXO and all its system components:
-
-            • Network configuration daemon
-            • Launch at login registration
-            • EXO network location
-
-            The app will be moved to Trash.
-            """
-        alert.alertStyle = .warning
-        alert.addButton(withTitle: "Uninstall")
-        alert.addButton(withTitle: "Cancel")
-
-        // Style the Uninstall button as destructive
-        if let uninstallButton = alert.buttons.first {
-            uninstallButton.hasDestructiveAction = true
-        }
-
-        let response = alert.runModal()
-        if response == .alertFirstButtonReturn {
-            performUninstall()
-        }
-    }
-
-    private func performUninstall() {
-        uninstallInProgress = true
-
-        // Stop EXO process first
-        controller.cancelPendingLaunch()
-        controller.stop()
-        stateService.stopPolling()
-
-        // Run the privileged uninstall on a background thread
-        // Using .utility QoS to avoid priority inversion with NSAppleScript's subprocess
-        DispatchQueue.global(qos: .utility).async {
-            do {
-                // Remove network setup daemon and components (requires admin privileges)
-                try NetworkSetupHelper.uninstall()
-
-                DispatchQueue.main.async {
-                    // Unregister from launch at login
-                    LaunchAtLoginHelper.disable()
-
-                    // Move app to trash
-                    self.moveAppToTrash()
-
-                    // Quit the app
-                    DispatchQueue.main.asyncAfter(deadline: .now() + 0.5) {
-                        NSApplication.shared.terminate(nil)
-                    }
-                }
-            } catch {
-                DispatchQueue.main.async {
-                    self.showErrorAlert(message: error.localizedDescription)
-                    self.uninstallInProgress = false
-                }
-            }
-        }
-    }
-
-    private func showErrorAlert(message: String) {
-        let alert = NSAlert()
-        alert.messageText = "Uninstall Failed"
-        alert.informativeText = message
-        alert.alertStyle = .critical
-        alert.addButton(withTitle: "OK")
-        alert.runModal()
-    }
-
-    private func moveAppToTrash() {
-        guard let appURL = Bundle.main.bundleURL as URL? else { return }
-        do {
-            try FileManager.default.trashItem(at: appURL, resultingItemURL: nil)
-        } catch {
-            // If we can't trash the app, that's OK - user can do it manually
-            // The important system components have already been cleaned up
-        }
-    }
-
     private var buildTag: String {
         Bundle.main.infoDictionary?["EXOBuildTag"] as? String ?? "unknown"
     }
@@ -568,24 +460,14 @@ private struct HoverButton: View {
     let title: String
     let tint: Color
     let trailingSystemImage: String?
-    let small: Bool
     let action: () -> Void
 
-    init(title: String, tint: Color = .primary, trailingSystemImage: String? = nil, small: Bool = false, action: @escaping () -> Void) {
-        self.title = title
-        self.tint = tint
-        self.trailingSystemImage = trailingSystemImage
-        self.small = small
-        self.action = action
-    }
-
     @State private var isHovering = false
 
     var body: some View {
         Button(action: action) {
             HStack {
                 Text(title)
-                    .font(small ? .caption : nil)
                 Spacer()
                 if let systemName = trailingSystemImage {
                     Image(systemName: systemName)
@@ -593,8 +475,8 @@ private struct HoverButton: View {
                 }
             }
             .frame(maxWidth: .infinity, alignment: .leading)
-            .padding(.vertical, small ? 4 : 6)
-            .padding(.horizontal, small ? 6 : 8)
+            .padding(.vertical, 6)
+            .padding(.horizontal, 8)
             .background(
                 RoundedRectangle(cornerRadius: 6)
                     .fill(

app/EXO/EXO/EXOApp.swift

@@ -125,22 +125,6 @@ struct EXOApp: App {
     }
 }
 
-/// Helper for managing EXO's launch-at-login registration
-enum LaunchAtLoginHelper {
-    private static let logger = Logger(subsystem: "io.exo.EXO", category: "LaunchAtLogin")
-
-    /// Unregisters EXO from launching at login
-    static func disable() {
-        guard SMAppService.mainApp.status == .enabled else { return }
-        do {
-            try SMAppService.mainApp.unregister()
-            logger.info("Unregistered EXO from launch at login")
-        } catch {
-            logger.error("Failed to unregister EXO from launch at login: \(error.localizedDescription, privacy: .public)")
-        }
-    }
-}
-
 final class SparkleUpdater: NSObject, ObservableObject {
     private let controller: SPUStandardUpdaterController
     private let delegateProxy: ExoUpdaterDelegate

app/EXO/EXO/Services/BugReportService.swift

@@ -82,6 +82,7 @@ struct BugReportService {
     }
 
     private func loadCredentials() throws -> AWSConfig {
+        // These credentials are write-only and necessary to receive bug reports from users
         return AWSConfig(
             accessKey: "AKIAYEKP5EMXTOBYDGHX",
             secretKey: "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",

app/EXO/EXO/Services/ClusterStateService.swift

@@ -7,7 +7,6 @@ final class ClusterStateService: ObservableObject {
     @Published private(set) var lastError: String?
     @Published private(set) var lastActionMessage: String?
     @Published private(set) var modelOptions: [ModelOption] = []
-    @Published private(set) var localNodeId: String?
 
     private var timer: Timer?
     private let decoder: JSONDecoder
@@ -30,7 +29,6 @@ final class ClusterStateService: ObservableObject {
     func startPolling(interval: TimeInterval = 0.5) {
         stopPolling()
         Task {
-            await fetchLocalNodeId()
             await fetchModels()
             await fetchSnapshot()
         }
@@ -48,31 +46,9 @@ final class ClusterStateService: ObservableObject {
         latestSnapshot = nil
         lastError = nil
         lastActionMessage = nil
-        localNodeId = nil
-    }
-
-    private func fetchLocalNodeId() async {
-        do {
-            let url = baseURL.appendingPathComponent("node_id")
-            var request = URLRequest(url: url)
-            request.cachePolicy = .reloadIgnoringLocalCacheData
-            let (data, response) = try await session.data(for: request)
-            guard let httpResponse = response as? HTTPURLResponse, (200..<300).contains(httpResponse.statusCode) else {
-                return
-            }
-            if let nodeId = try? decoder.decode(String.self, from: data) {
-                localNodeId = nodeId
-            }
-        } catch {
-            // Silently ignore - localNodeId will remain nil and retry on next poll
-        }
     }
 
     private func fetchSnapshot() async {
-        // Retry fetching local node ID if not yet set
-        if localNodeId == nil {
-            await fetchLocalNodeId()
-        }
         do {
             var request = URLRequest(url: endpoint)
             request.cachePolicy = .reloadIgnoringLocalCacheData

app/EXO/EXO/Services/NetworkSetupHelper.swift

@@ -62,8 +62,7 @@ networksetup -listnetworkservices | grep -q "Thunderbolt Bridge" && {
 """
 
     static func ensureLaunchDaemonInstalled() {
-        // Use .utility priority to match NSAppleScript's internal QoS and avoid priority inversion
-        Task.detached(priority: .utility) {
+        Task.detached {
             do {
                 if daemonAlreadyInstalled() {
                     return
@@ -76,63 +75,6 @@ networksetup -listnetworkservices | grep -q "Thunderbolt Bridge" && {
         }
     }
 
-    /// Removes all EXO network setup components from the system.
-    /// This includes the LaunchDaemon, scripts, logs, and network location.
-    /// Requires admin privileges.
-    static func uninstall() throws {
-        let uninstallScript = makeUninstallScript()
-        try runShellAsAdmin(uninstallScript)
-        logger.info("EXO network setup components removed successfully")
-    }
-
-    /// Checks if there are any EXO network components installed that need cleanup
-    static func hasInstalledComponents() -> Bool {
-        let manager = FileManager.default
-        let scriptExists = manager.fileExists(atPath: scriptDestination)
-        let plistExists = manager.fileExists(atPath: plistDestination)
-        return scriptExists || plistExists
-    }
-
-    private static func makeUninstallScript() -> String {
-        """
-set -euo pipefail
-
-LABEL="\(daemonLabel)"
-SCRIPT_DEST="\(scriptDestination)"
-PLIST_DEST="\(plistDestination)"
-LOG_OUT="/var/log/\(daemonLabel).log"
-LOG_ERR="/var/log/\(daemonLabel).err.log"
-
-# Unload the LaunchDaemon if running
-launchctl bootout system/"$LABEL" 2>/dev/null || true
-
-# Remove LaunchDaemon plist
-rm -f "$PLIST_DEST"
-
-# Remove the script and parent directory if empty
-rm -f "$SCRIPT_DEST"
-rmdir "$(dirname "$SCRIPT_DEST")" 2>/dev/null || true
-
-# Remove log files
-rm -f "$LOG_OUT" "$LOG_ERR"
-
-# Switch back to Automatic network location
-networksetup -switchtolocation Automatic 2>/dev/null || true
-
-# Delete the exo network location if it exists
-networksetup -listlocations | grep -q '^exo$' && {
-  networksetup -deletelocation exo 2>/dev/null || true
-} || true
-
-# Re-enable Thunderbolt Bridge if it exists
-networksetup -listnetworkservices | grep -q "Thunderbolt Bridge" && {
-  networksetup -setnetworkserviceenabled "Thunderbolt Bridge" on 2>/dev/null || true
-} || true
-
-echo "EXO network components removed successfully"
-"""
-    }
-
     private static func daemonAlreadyInstalled() -> Bool {
         let manager = FileManager.default
         let scriptExists = manager.fileExists(atPath: scriptDestination)

app/EXO/EXO/ViewModels/NodeViewModel.swift

@@ -85,7 +85,7 @@ struct TopologyViewModel {
 }
 
 extension ClusterState {
-    func topologyViewModel(localNodeId: String?) -> TopologyViewModel? {
+    func topologyViewModel() -> TopologyViewModel? {
         let topologyNodeIds = Set(topology?.nodes.map(\.nodeId) ?? [])
         let allNodes = nodeViewModels().filter { topologyNodeIds.isEmpty || topologyNodeIds.contains($0.id) }
         guard !allNodes.isEmpty else { return nil }
@@ -105,11 +105,6 @@ extension ClusterState {
             orderedNodes = allNodes
         }
 
-        // Rotate so the local node (from /node_id API) is first
-        if let localId = localNodeId, let index = orderedNodes.firstIndex(where: { $0.id == localId }) {
-            orderedNodes = Array(orderedNodes[index...]) + Array(orderedNodes[..<index])
-        }
-
         let nodeIds = Set(orderedNodes.map(\.id))
         let edgesArray: [TopologyEdgeViewModel] = topology?.connections?.compactMap { connection in
             guard nodeIds.contains(connection.localNodeId), nodeIds.contains(connection.sendBackNodeId) else { return nil }
@@ -117,7 +112,10 @@ extension ClusterState {
         } ?? []
         let edges = Set(edgesArray)
 
-        return TopologyViewModel(nodes: orderedNodes, edges: Array(edges), currentNodeId: localNodeId)
+        let topologyRootId = topology?.nodes.first?.nodeId
+        let currentId = orderedNodes.first(where: { $0.id == topologyRootId })?.id ?? orderedNodes.first?.id
+
+        return TopologyViewModel(nodes: orderedNodes, edges: Array(edges), currentNodeId: currentId)
     }
 }
 

app/EXO/uninstall-exo.sh

@@ -1,154 +0,0 @@
-#!/usr/bin/env bash
-#
-# EXO Uninstaller Script
-#
-# This script removes all EXO system components that persist after deleting the app.
-# Run with: sudo ./uninstall-exo.sh
-#
-# Components removed:
-# - LaunchDaemon: /Library/LaunchDaemons/io.exo.networksetup.plist
-# - Network script: /Library/Application Support/EXO/
-# - Log files: /var/log/io.exo.networksetup.*
-# - Network location: "exo"
-# - Launch at login registration
-#
-
-set -euo pipefail
-
-LABEL="io.exo.networksetup"
-SCRIPT_DEST="/Library/Application Support/EXO/disable_bridge_enable_dhcp.sh"
-PLIST_DEST="/Library/LaunchDaemons/io.exo.networksetup.plist"
-LOG_OUT="/var/log/${LABEL}.log"
-LOG_ERR="/var/log/${LABEL}.err.log"
-APP_BUNDLE_ID="io.exo.EXO"
-
-# Colors for output
-RED='\033[0;31m'
-GREEN='\033[0;32m'
-YELLOW='\033[1;33m'
-NC='\033[0m' # No Color
-
-echo_info() {
-    echo -e "${GREEN}[INFO]${NC} $1"
-}
-
-echo_warn() {
-    echo -e "${YELLOW}[WARN]${NC} $1"
-}
-
-echo_error() {
-    echo -e "${RED}[ERROR]${NC} $1"
-}
-
-# Check if running as root
-if [[ $EUID -ne 0 ]]; then
-    echo_error "This script must be run as root (use sudo)"
-    exit 1
-fi
-
-echo ""
-echo "========================================"
-echo "        EXO Uninstaller"
-echo "========================================"
-echo ""
-
-# Unload the LaunchDaemon if running
-echo_info "Stopping network setup daemon..."
-if launchctl list | grep -q "$LABEL"; then
-    launchctl bootout system/"$LABEL" 2>/dev/null || true
-    echo_info "Daemon stopped"
-else
-    echo_warn "Daemon was not running"
-fi
-
-# Remove LaunchDaemon plist
-if [[ -f "$PLIST_DEST" ]]; then
-    rm -f "$PLIST_DEST"
-    echo_info "Removed LaunchDaemon plist"
-else
-    echo_warn "LaunchDaemon plist not found (already removed?)"
-fi
-
-# Remove the script and parent directory
-if [[ -f "$SCRIPT_DEST" ]]; then
-    rm -f "$SCRIPT_DEST"
-    echo_info "Removed network setup script"
-else
-    echo_warn "Network setup script not found (already removed?)"
-fi
-
-# Remove EXO directory if empty
-if [[ -d "/Library/Application Support/EXO" ]]; then
-    rmdir "/Library/Application Support/EXO" 2>/dev/null && \
-        echo_info "Removed EXO support directory" || \
-        echo_warn "EXO support directory not empty, leaving in place"
-fi
-
-# Remove log files
-if [[ -f "$LOG_OUT" ]] || [[ -f "$LOG_ERR" ]]; then
-    rm -f "$LOG_OUT" "$LOG_ERR"
-    echo_info "Removed log files"
-else
-    echo_warn "Log files not found (already removed?)"
-fi
-
-# Switch back to Automatic network location
-echo_info "Restoring network configuration..."
-if networksetup -listlocations | grep -q "^Automatic$"; then
-    networksetup -switchtolocation Automatic 2>/dev/null || true
-    echo_info "Switched to Automatic network location"
-else
-    echo_warn "Automatic network location not found"
-fi
-
-# Delete the exo network location if it exists
-if networksetup -listlocations | grep -q "^exo$"; then
-    networksetup -deletelocation exo 2>/dev/null || true
-    echo_info "Deleted 'exo' network location"
-else
-    echo_warn "'exo' network location not found (already removed?)"
-fi
-
-# Re-enable Thunderbolt Bridge if it exists
-if networksetup -listnetworkservices 2>/dev/null | grep -q "Thunderbolt Bridge"; then
-    networksetup -setnetworkserviceenabled "Thunderbolt Bridge" on 2>/dev/null || true
-    echo_info "Re-enabled Thunderbolt Bridge"
-fi
-
-# Note about launch at login registration
-# SMAppService-based login items cannot be removed from a shell script.
-# They can only be unregistered from within the app itself or manually via System Settings.
-echo_warn "Launch at login must be removed manually:"
-echo_warn "  System Settings → General → Login Items → Remove EXO"
-
-# Check if EXO.app exists in common locations
-APP_FOUND=false
-for app_path in "/Applications/EXO.app" "$HOME/Applications/EXO.app"; do
-    if [[ -d "$app_path" ]]; then
-        if [[ "$APP_FOUND" == false ]]; then
-            echo ""
-            APP_FOUND=true
-        fi
-        echo_warn "EXO.app found at: $app_path"
-        echo_warn "You may want to move it to Trash manually."
-    fi
-done
-
-echo ""
-echo "========================================"
-echo_info "EXO uninstall complete!"
-echo "========================================"
-echo ""
-echo "The following have been removed:"
-echo "  • Network setup LaunchDaemon"
-echo "  • Network configuration script"
-echo "  • Log files"
-echo "  • 'exo' network location"
-echo ""
-echo "Your network has been restored to use the 'Automatic' location."
-echo "Thunderbolt Bridge has been re-enabled (if present)."
-echo ""
-echo "Manual step required:"
-echo "  Remove EXO from Login Items in System Settings → General → Login Items"
-echo ""
-

dashboard/package-lock.json

@@ -9,8 +9,6 @@
 			"version": "1.0.0",
 			"dependencies": {
 				"highlight.js": "^11.11.1",
-				"katex": "^0.16.27",
-				"marked": "^17.0.1",
 				"mode-watcher": "^1.1.0"
 			},
 			"devDependencies": {
@@ -863,6 +861,7 @@
 			"integrity": "sha512-oH8tXw7EZnie8FdOWYrF7Yn4IKrqTFHhXvl8YxXxbKwTMcD/5NNCryUSEXRk2ZR4ojnub0P8rNrsVGHXWqIDtA==",
 			"dev": true,
 			"license": "MIT",
+			"peer": true,
 			"dependencies": {
 				"@standard-schema/spec": "^1.0.0",
 				"@sveltejs/acorn-typescript": "^1.0.5",
@@ -902,6 +901,7 @@
 			"integrity": "sha512-Y1Cs7hhTc+a5E9Va/xwKlAJoariQyHY+5zBgCZg4PFWNYQ1nMN9sjK1zhw1gK69DuqVP++sht/1GZg1aRwmAXQ==",
 			"dev": true,
 			"license": "MIT",
+			"peer": true,
 			"dependencies": {
 				"@sveltejs/vite-plugin-svelte-inspector": "^4.0.1",
 				"debug": "^4.4.1",
@@ -1518,6 +1518,7 @@
 			"integrity": "sha512-LCCV0HdSZZZb34qifBsyWlUmok6W7ouER+oQIGBScS8EsZsQbrtFTUrDX4hOl+CS6p7cnNC4td+qrSVGSCTUfQ==",
 			"dev": true,
 			"license": "MIT",
+			"peer": true,
 			"dependencies": {
 				"undici-types": "~6.21.0"
 			}
@@ -1527,6 +1528,7 @@
 			"resolved": "https://registry.npmjs.org/acorn/-/acorn-8.15.0.tgz",
 			"integrity": "sha512-NZyJarBfL7nWwIq+FDL6Zp/yHEhePMNnnJ0y3qfieCrmNvYct8uvtiV41UvlSe6apAfk0fY1FbWx+NwfmpvtTg==",
 			"license": "MIT",
+			"peer": true,
 			"bin": {
 				"acorn": "bin/acorn"
 			},
@@ -1939,6 +1941,7 @@
 			"integrity": "sha512-fmTRWbNMmsmWq6xJV8D19U/gw/bwrHfNXxrIN+HfZgnzqTHp9jOmKMhsTUjXOJnZOdZY9Q28y4yebKzqDKlxlQ==",
 			"dev": true,
 			"license": "ISC",
+			"peer": true,
 			"engines": {
 				"node": ">=12"
 			}
@@ -2251,31 +2254,6 @@
 				"jiti": "lib/jiti-cli.mjs"
 			}
 		},
-		"node_modules/katex": {
-			"version": "0.16.27",
-			"resolved": "https://registry.npmjs.org/katex/-/katex-0.16.27.tgz",
-			"integrity": "sha512-aeQoDkuRWSqQN6nSvVCEFvfXdqo1OQiCmmW1kc9xSdjutPv7BGO7pqY9sQRJpMOGrEdfDgF2TfRXe5eUAD2Waw==",
-			"funding": [
-				"https://opencollective.com/katex",
-				"https://github.com/sponsors/katex"
-			],
-			"license": "MIT",
-			"dependencies": {
-				"commander": "^8.3.0"
-			},
-			"bin": {
-				"katex": "cli.js"
-			}
-		},
-		"node_modules/katex/node_modules/commander": {
-			"version": "8.3.0",
-			"resolved": "https://registry.npmjs.org/commander/-/commander-8.3.0.tgz",
-			"integrity": "sha512-OkTL9umf+He2DZkUq8f8J9of7yL6RJKI24dVITBmNfZBmri9zYZQrKkuXiKhyfPSu8tUhnVBB1iKXevvnlR4Ww==",
-			"license": "MIT",
-			"engines": {
-				"node": ">= 12"
-			}
-		},
 		"node_modules/kleur": {
 			"version": "4.1.5",
 			"resolved": "https://registry.npmjs.org/kleur/-/kleur-4.1.5.tgz",
@@ -2562,18 +2540,6 @@
 				"@jridgewell/sourcemap-codec": "^1.5.5"
 			}
 		},
-		"node_modules/marked": {
-			"version": "17.0.1",
-			"resolved": "https://registry.npmjs.org/marked/-/marked-17.0.1.tgz",
-			"integrity": "sha512-boeBdiS0ghpWcSwoNm/jJBwdpFaMnZWRzjA6SkUMYb40SVaN1x7mmfGKp0jvexGcx+7y2La5zRZsYFZI6Qpypg==",
-			"license": "MIT",
-			"bin": {
-				"marked": "bin/marked.js"
-			},
-			"engines": {
-				"node": ">= 20"
-			}
-		},
 		"node_modules/mode-watcher": {
 			"version": "1.1.0",
 			"resolved": "https://registry.npmjs.org/mode-watcher/-/mode-watcher-1.1.0.tgz",
@@ -2646,6 +2612,7 @@
 			"integrity": "sha512-5gTmgEY/sqK6gFXLIsQNH19lWb4ebPDLA4SdLP7dsWkIXHWlG66oPuVvXSGFPppYZz8ZDZq0dYYrbHfBCVUb1Q==",
 			"dev": true,
 			"license": "MIT",
+			"peer": true,
 			"engines": {
 				"node": ">=12"
 			},
@@ -2833,6 +2800,7 @@
 			"resolved": "https://registry.npmjs.org/svelte/-/svelte-5.45.3.tgz",
 			"integrity": "sha512-ngKXNhNvwPzF43QqEhDOue7TQTrG09em1sd4HBxVF0Wr2gopAmdEWan+rgbdgK4fhBtSOTJO8bYU4chUG7VXZQ==",
 			"license": "MIT",
+			"peer": true,
 			"dependencies": {
 				"@jridgewell/remapping": "^2.3.4",
 				"@jridgewell/sourcemap-codec": "^1.5.0",
@@ -2977,6 +2945,7 @@
 			"integrity": "sha512-jl1vZzPDinLr9eUt3J/t7V6FgNEw9QjvBPdysz9KfQDD41fQrC2Y4vKQdiaUpFT4bXlb1RHhLpp8wtm6M5TgSw==",
 			"dev": true,
 			"license": "Apache-2.0",
+			"peer": true,
 			"bin": {
 				"tsc": "bin/tsc",
 				"tsserver": "bin/tsserver"
@@ -2998,6 +2967,7 @@
 			"integrity": "sha512-+Oxm7q9hDoLMyJOYfUYBuHQo+dkAloi33apOPP56pzj+vsdJDzr+j1NISE5pyaAuKL4A3UD34qd0lx5+kfKp2g==",
 			"dev": true,
 			"license": "MIT",
+			"peer": true,
 			"dependencies": {
 				"esbuild": "^0.25.0",
 				"fdir": "^6.4.4",

dashboard/package.json

@@ -27,8 +27,7 @@
 	},
 	"dependencies": {
 		"highlight.js": "^11.11.1",
-		"katex": "^0.16.27",
-		"marked": "^17.0.1",
 		"mode-watcher": "^1.1.0"
 	}
 }
+

dashboard/src/lib/components/ChatForm.svelte

@@ -139,11 +139,6 @@
 	}
 
 	function handleKeydown(event: KeyboardEvent) {
-		// Prevent form submission during IME composition (e.g., Chinese, Japanese, Korean input)
-		if (event.isComposing || event.keyCode === 229) {
-			return;
-		}
-		
 		if (event.key === 'Enter' && !event.shiftKey) {
 			event.preventDefault();
 			handleSubmit();

dashboard/src/lib/components/ChatMessages.svelte

@@ -8,80 +8,89 @@
 		regenerateLastResponse
 	} from '$lib/stores/app.svelte';
 	import type { MessageAttachment } from '$lib/stores/app.svelte';
-	import MarkdownContent from './MarkdownContent.svelte';
+import { tick, onDestroy } from 'svelte';
 
-	interface Props {
-		class?: string;
-		scrollParent?: HTMLElement | null;
-	}
+interface Props {
+	class?: string;
+	scrollParent?: HTMLElement | null;
+}
 
-	let { class: className = '', scrollParent = null }: Props = $props();
+let { class: className = '', scrollParent = null }: Props = $props();
 
 	const messageList = $derived(messages());
 	const response = $derived(currentResponse());
 	const loading = $derived(isLoading());
 
-	// Scroll management - user controls scroll, show button when not at bottom
-	const SCROLL_THRESHOLD = 100;
-	let showScrollButton = $state(false);
-	let lastMessageCount = 0;
-	let containerRef: HTMLDivElement | undefined = $state();
+// Ref for scroll anchor at bottom
+let scrollAnchorRef: HTMLDivElement | undefined = $state();
 
-	function getScrollContainer(): HTMLElement | null {
-		if (scrollParent) return scrollParent;
-		return containerRef?.parentElement ?? null;
+// Scroll management
+const SCROLL_BOTTOM_THRESHOLD = 120;
+let autoScrollEnabled = true;
+let currentScrollEl: HTMLElement | null = null;
+
+function resolveScrollElement(): HTMLElement | null {
+	if (scrollParent) return scrollParent;
+	let node: HTMLElement | null = scrollAnchorRef?.parentElement as HTMLElement | null;
+	while (node) {
+		const isScrollable = node.scrollHeight > node.clientHeight + 1;
+		if (isScrollable) return node;
+		node = node.parentElement;
 	}
+	return null;
+}
 
-	function isNearBottom(el: HTMLElement): boolean {
-		return el.scrollHeight - el.scrollTop - el.clientHeight < SCROLL_THRESHOLD;
+function handleScroll() {
+	if (!currentScrollEl) return;
+	const distanceFromBottom = currentScrollEl.scrollHeight - currentScrollEl.scrollTop - currentScrollEl.clientHeight;
+	const isNearBottom = distanceFromBottom < SCROLL_BOTTOM_THRESHOLD;
+	autoScrollEnabled = isNearBottom;
+}
+
+function attachScrollListener() {
+	const nextEl = resolveScrollElement();
+	if (currentScrollEl === nextEl) return;
+	if (currentScrollEl) {
+		currentScrollEl.removeEventListener('scroll', handleScroll);
 	}
+	currentScrollEl = nextEl;
+	if (currentScrollEl) {
+		currentScrollEl.addEventListener('scroll', handleScroll);
+		// Initialize state based on current position
+		handleScroll();
+	}
+}
 
-	function scrollToBottom() {
-		const el = getScrollContainer();
-		if (el) {
-			el.scrollTo({ top: el.scrollHeight, behavior: 'smooth' });
+onDestroy(() => {
+	if (currentScrollEl) {
+		currentScrollEl.removeEventListener('scroll', handleScroll);
+	}
+});
+
+$effect(() => {
+	// Re-evaluate scroll container if prop changes or after mount
+	scrollParent;
+	attachScrollListener();
+});
+
+// Auto-scroll to bottom when messages change or response updates, but only if user is near bottom
+$effect(() => {
+	// Track these values to trigger effect
+	const _ = messageList.length;
+	const __ = response;
+	const ___ = loading;
+	
+	tick().then(() => {
+		const el = currentScrollEl ?? resolveScrollElement();
+		if (!el || !scrollAnchorRef) return;
+		const distanceFromBottom = el.scrollHeight - el.scrollTop - el.clientHeight;
+		const isNearBottom = distanceFromBottom < SCROLL_BOTTOM_THRESHOLD;
+		if (autoScrollEnabled || isNearBottom) {
+			scrollAnchorRef.scrollIntoView({ behavior: 'smooth', block: 'end' });
+			autoScrollEnabled = true;
 		}
-	}
-
-	function updateScrollButtonVisibility() {
-		const el = getScrollContainer();
-		if (!el) return;
-		showScrollButton = !isNearBottom(el);
-	}
-
-	// Attach scroll listener
-	$effect(() => {
-		const el = scrollParent ?? containerRef?.parentElement;
-		if (!el) return;
-		
-		el.addEventListener('scroll', updateScrollButtonVisibility, { passive: true });
-		// Initial check
-		updateScrollButtonVisibility();
-		return () => el.removeEventListener('scroll', updateScrollButtonVisibility);
-	});
-
-	// Auto-scroll when user sends a new message
-	$effect(() => {
-		const count = messageList.length;
-		if (count > lastMessageCount) {
-			const el = getScrollContainer();
-			if (el) {
-				requestAnimationFrame(() => {
-					el.scrollTo({ top: el.scrollHeight, behavior: 'smooth' });
-				});
-			}
-		}
-		lastMessageCount = count;
-	});
-
-	// Update scroll button visibility when content changes
-	$effect(() => {
-		// Track response to trigger re-check during streaming
-		const _ = response;
-		
-		// Small delay to let DOM update
-		requestAnimationFrame(() => updateScrollButtonVisibility());
 	});
+});
 
 	// Edit state
 	let editingMessageId = $state<string | null>(null);
@@ -222,7 +231,7 @@ function isThinkingExpanded(messageId: string): boolean {
 <div class="flex flex-col gap-4 sm:gap-6 {className}">
 	{#each messageList as message (message.id)}
 		<div class="group flex {message.role === 'user' ? 'justify-end' : 'justify-start'}">
-			<div class="{message.role === 'user' ? 'max-w-[85%] sm:max-w-[70%] flex flex-col items-end' : 'w-full max-w-[98%] sm:max-w-[95%]'}">
+			<div class="{message.role === 'user' ? 'max-w-[85%] sm:max-w-[70%] flex flex-col items-end' : 'max-w-[95%] sm:max-w-[85%]'}">
 				{#if message.role === 'assistant'}
 					<!-- Assistant message header -->
 					<div class="flex items-center gap-1.5 sm:gap-2 mb-1.5 sm:mb-2">
@@ -296,7 +305,7 @@ function isThinkingExpanded(messageId: string): boolean {
 				{:else}
 					<div class="{message.role === 'user' 
 						? 'command-panel rounded-lg rounded-tr-sm inline-block' 
-						: 'command-panel rounded-lg rounded-tl-sm border-l-2 border-l-exo-yellow/50 block w-full'}">
+						: 'command-panel rounded-lg rounded-tl-sm border-l-2 border-l-exo-yellow/50 inline-block'}">
 						
 						{#if message.role === 'user'}
 							<!-- User message styling -->
@@ -322,7 +331,7 @@ function isThinkingExpanded(messageId: string): boolean {
 								{/if}
 								
 								{#if message.content}
-									<div class="text-xs text-foreground font-mono tracking-wide whitespace-pre-wrap break-words leading-relaxed">
+									<div class="text-sm text-foreground font-mono tracking-wide whitespace-pre-wrap break-words leading-relaxed">
 										{message.content}
 									</div>
 								{/if}
@@ -351,7 +360,7 @@ function isThinkingExpanded(messageId: string): boolean {
 												</svg>
 												<span>Thinking...</span>
 											</span>
-											<span class="text-[10px] tracking-[0.2em] text-exo-light-gray/60 ml-4">
+											<span class="text-[10px] tracking-[0.2em] text-exo-light-gray/60">
 												{isThinkingExpanded(message.id) ? 'HIDE' : 'SHOW'}
 											</span>
 										</button>
@@ -365,8 +374,8 @@ function isThinkingExpanded(messageId: string): boolean {
 										{/if}
 									</div>
 								{/if}
-								<div class="text-xs text-foreground">
-									<MarkdownContent content={message.content || (loading ? response : '')} />
+								<div class="text-sm text-foreground font-mono tracking-wide whitespace-pre-wrap break-words leading-relaxed">
+									{message.content || (loading ? response : '')}
 									{#if loading && !message.content}
 										<span class="inline-block w-2 h-4 bg-exo-yellow/70 ml-1 cursor-blink"></span>
 									{/if}
@@ -448,20 +457,6 @@ function isThinkingExpanded(messageId: string): boolean {
 		</div>
 	{/if}
 	
-	<!-- Invisible element for container reference -->
-	<div bind:this={containerRef}></div>
-
-	<!-- Scroll to bottom button -->
-	{#if showScrollButton}
-		<button
-			type="button"
-			onclick={scrollToBottom}
-			class="sticky bottom-4 left-1/2 -translate-x-1/2 w-10 h-10 rounded-full bg-exo-dark-gray/90 border border-exo-medium-gray/50 flex items-center justify-center text-exo-light-gray hover:text-exo-yellow hover:border-exo-yellow/50 transition-all shadow-lg cursor-pointer z-10"
-			title="Scroll to bottom"
-		>
-			<svg class="w-5 h-5" fill="none" viewBox="0 0 24 24" stroke="currentColor">
-				<path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M19 14l-7 7m0 0l-7-7m7 7V3" />
-			</svg>
-		</button>
-	{/if}
+	<!-- Scroll anchor for auto-scroll -->
+	<div bind:this={scrollAnchorRef}></div>
 </div>

dashboard/src/lib/components/ChatSidebar.svelte

@@ -10,9 +10,7 @@ import {
 		clearChat,
 		instances,
 		debugMode,
-		toggleDebugMode,
-		topologyOnlyMode,
-		toggleTopologyOnlyMode
+		toggleDebugMode
 	} from '$lib/stores/app.svelte';
 
 	interface Props {
@@ -25,7 +23,6 @@ import {
 	const activeId = $derived(activeConversationId());
 const instanceData = $derived(instances());
 const debugEnabled = $derived(debugMode());
-const topologyOnlyEnabled = $derived(topologyOnlyMode());
 
 	let searchQuery = $state('');
 	let editingId = $state<string | null>(null);
@@ -427,19 +424,6 @@ const topologyOnlyEnabled = $derived(topologyOnlyMode());
 		<div class="text-xs text-white/60 font-mono tracking-wider text-center">
 			{conversationList.length} CONVERSATION{conversationList.length !== 1 ? 'S' : ''}
 		</div>
-		<button
-			type="button"
-			onclick={toggleTopologyOnlyMode}
-			class="p-1.5 rounded border border-exo-medium-gray/40 hover:border-exo-yellow/50 transition-colors cursor-pointer"
-			title="Toggle topology only mode"
-		>
-			<svg class="w-4 h-4 {topologyOnlyEnabled ? 'text-exo-yellow' : 'text-exo-medium-gray'}" fill="none" viewBox="0 0 24 24" stroke="currentColor" stroke-width="2">
-				<circle cx="12" cy="5" r="2" fill="currentColor" />
-				<circle cx="5" cy="19" r="2" fill="currentColor" />
-				<circle cx="19" cy="19" r="2" fill="currentColor" />
-				<path stroke-linecap="round" d="M12 7v5m0 0l-5 5m5-5l5 5" />
-			</svg>
-		</button>
 	</div>
 	</div>
 </aside>

dashboard/src/lib/components/HeaderNav.svelte

@@ -3,9 +3,6 @@
 
 	export let showHome = true;
 	export let onHome: (() => void) | null = null;
-	export let showSidebarToggle = false;
-	export let sidebarVisible = true;
-	export let onToggleSidebar: (() => void) | null = null;
 
 	function handleHome(): void {
 		if (onHome) {
@@ -17,38 +14,13 @@
 			window.location.hash = '/';
 		}
 	}
-
-	function handleToggleSidebar(): void {
-		if (onToggleSidebar) {
-			onToggleSidebar();
-		}
-	}
 </script>
 
 <header class="relative z-20 flex items-center justify-center px-6 pt-8 pb-4 bg-exo-dark-gray">
-	<!-- Left: Sidebar Toggle -->
-	{#if showSidebarToggle}
-	<div class="absolute left-6 top-1/2 -translate-y-1/2">
-		<button
-			onclick={handleToggleSidebar}
-			class="p-2 rounded border border-exo-medium-gray/40 hover:border-exo-yellow/50 transition-colors cursor-pointer"
-			title={sidebarVisible ? 'Hide sidebar' : 'Show sidebar'}
-		>
-			<svg class="w-5 h-5 {sidebarVisible ? 'text-exo-yellow' : 'text-exo-medium-gray'}" fill="none" viewBox="0 0 24 24" stroke="currentColor" stroke-width="2">
-				{#if sidebarVisible}
-					<path stroke-linecap="round" stroke-linejoin="round" d="M11 19l-7-7 7-7m8 14l-7-7 7-7" />
-				{:else}
-					<path stroke-linecap="round" stroke-linejoin="round" d="M13 5l7 7-7 7M5 5l7 7-7 7" />
-				{/if}
-			</svg>
-		</button>
-	</div>
-	{/if}
-
 	<!-- Center: Logo (clickable to go home) -->
 	<button
 		onclick={handleHome}
-		class="bg-transparent border-none outline-none focus:outline-none transition-opacity duration-200 hover:opacity-90 {showHome ? 'cursor-pointer' : 'cursor-default'}"
+		class="hover:opacity-80 transition-opacity {showHome ? 'cursor-pointer' : 'cursor-default'}"
 		title={showHome ? 'Go to home' : ''}
 		disabled={!showHome}
 	>

dashboard/src/lib/components/MarkdownContent.svelte

@@ -1,451 +0,0 @@
-<script lang="ts">
-	import { marked } from 'marked';
-	import hljs from 'highlight.js';
-	import katex from 'katex';
-	import 'katex/dist/katex.min.css';
-	import { browser } from '$app/environment';
-
-	interface Props {
-		content: string;
-		class?: string;
-	}
-
-	let { content, class: className = '' }: Props = $props();
-
-	let containerRef = $state<HTMLDivElement>();
-	let processedHtml = $state('');
-
-	// Configure marked with syntax highlighting
-	marked.setOptions({
-		gfm: true,
-		breaks: true
-	});
-
-	// Custom renderer for code blocks
-	const renderer = new marked.Renderer();
-
-	renderer.code = function ({ text, lang }: { text: string; lang?: string }) {
-		const language = lang && hljs.getLanguage(lang) ? lang : 'plaintext';
-		const highlighted = hljs.highlight(text, { language }).value;
-		const codeId = `code-${Date.now()}-${Math.random().toString(36).slice(2, 9)}`;
-
-		return `
-			<div class="code-block-wrapper">
-				<div class="code-block-header">
-					<span class="code-language">${language}</span>
-					<button type="button" class="copy-code-btn" data-code="${encodeURIComponent(text)}" title="Copy code">
-						<svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
-							<rect width="14" height="14" x="8" y="8" rx="2" ry="2"/>
-							<path d="M4 16c-1.1 0-2-.9-2-2V4c0-1.1.9-2 2-2h10c1.1 0 2 .9 2 2"/>
-						</svg>
-					</button>
-				</div>
-				<pre><code class="hljs language-${language}" data-code-id="${codeId}">${highlighted}</code></pre>
-			</div>
-		`;
-	};
-
-	// Inline code
-	renderer.codespan = function ({ text }: { text: string }) {
-		return `<code class="inline-code">${text}</code>`;
-	};
-
-	marked.use({ renderer });
-
-	/**
-	 * Preprocess LaTeX: convert \(...\) to $...$ and \[...\] to $$...$$
-	 * Also protect code blocks from LaTeX processing
-	 */
-	function preprocessLaTeX(text: string): string {
-		// Protect code blocks
-		const codeBlocks: string[] = [];
-		let processed = text.replace(/```[\s\S]*?```|`[^`]+`/g, (match) => {
-			codeBlocks.push(match);
-			return `<<CODE_${codeBlocks.length - 1}>>`;
-		});
-
-		// Convert \(...\) to $...$
-		processed = processed.replace(/\\\((.+?)\\\)/g, '$$$1$');
-		
-		// Convert \[...\] to $$...$$
-		processed = processed.replace(/\\\[([\s\S]*?)\\\]/g, '$$$$$1$$$$');
-
-		// Restore code blocks
-		processed = processed.replace(/<<CODE_(\d+)>>/g, (_, index) => codeBlocks[parseInt(index)]);
-
-		return processed;
-	}
-
-	/**
-	 * Render math expressions with KaTeX after HTML is generated
-	 */
-	function renderMath(html: string): string {
-		// Render display math ($$...$$)
-		html = html.replace(/\$\$([\s\S]*?)\$\$/g, (_, math) => {
-			try {
-				return katex.renderToString(math.trim(), {
-					displayMode: true,
-					throwOnError: false,
-					output: 'html'
-				});
-			} catch {
-				return `<span class="math-error">$$${math}$$</span>`;
-			}
-		});
-
-		// Render inline math ($...$) but avoid matching currency like $5
-		html = html.replace(/\$([^\$\n]+?)\$/g, (match, math) => {
-			// Skip if it looks like currency ($ followed by number)
-			if (/^\d/.test(math.trim())) {
-				return match;
-			}
-			try {
-				return katex.renderToString(math.trim(), {
-					displayMode: false,
-					throwOnError: false,
-					output: 'html'
-				});
-			} catch {
-				return `<span class="math-error">$${math}$</span>`;
-			}
-		});
-
-		return html;
-	}
-
-	function processMarkdown(text: string): string {
-		try {
-			// Preprocess LaTeX notation
-			const preprocessed = preprocessLaTeX(text);
-			// Parse markdown
-			let html = marked.parse(preprocessed) as string;
-			// Render math expressions
-			html = renderMath(html);
-			return html;
-		} catch (error) {
-			console.error('Markdown processing error:', error);
-			return text.replace(/\n/g, '<br>');
-		}
-	}
-
-	async function handleCopyClick(event: Event) {
-		const target = event.currentTarget as HTMLButtonElement;
-		const encodedCode = target.getAttribute('data-code');
-		if (!encodedCode) return;
-
-		const code = decodeURIComponent(encodedCode);
-
-		try {
-			await navigator.clipboard.writeText(code);
-			// Show copied feedback
-			const originalHtml = target.innerHTML;
-			target.innerHTML = `
-				<svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
-					<path d="M20 6L9 17l-5-5"/>
-				</svg>
-			`;
-			target.classList.add('copied');
-			setTimeout(() => {
-				target.innerHTML = originalHtml;
-				target.classList.remove('copied');
-			}, 2000);
-		} catch (error) {
-			console.error('Failed to copy:', error);
-		}
-	}
-
-	function setupCopyButtons() {
-		if (!containerRef || !browser) return;
-
-		const buttons = containerRef.querySelectorAll<HTMLButtonElement>('.copy-code-btn');
-		for (const button of buttons) {
-			if (button.dataset.listenerBound !== 'true') {
-				button.dataset.listenerBound = 'true';
-				button.addEventListener('click', handleCopyClick);
-			}
-		}
-	}
-
-	$effect(() => {
-		if (content) {
-			processedHtml = processMarkdown(content);
-		} else {
-			processedHtml = '';
-		}
-	});
-
-	$effect(() => {
-		if (containerRef && processedHtml) {
-			setupCopyButtons();
-		}
-	});
-</script>
-
-<div bind:this={containerRef} class="markdown-content {className}">
-	{@html processedHtml}
-</div>
-
-<style>
-	.markdown-content {
-		line-height: 1.6;
-	}
-
-	/* Paragraphs */
-	.markdown-content :global(p) {
-		margin-bottom: 1rem;
-	}
-
-	.markdown-content :global(p:last-child) {
-		margin-bottom: 0;
-	}
-
-	/* Headers */
-	.markdown-content :global(h1) {
-		font-size: 1.5rem;
-		font-weight: 700;
-		margin: 1.5rem 0 0.75rem 0;
-		color: var(--exo-yellow, #ffd700);
-	}
-
-	.markdown-content :global(h2) {
-		font-size: 1.25rem;
-		font-weight: 600;
-		margin: 1.25rem 0 0.5rem 0;
-		color: var(--exo-yellow, #ffd700);
-	}
-
-	.markdown-content :global(h3) {
-		font-size: 1.125rem;
-		font-weight: 600;
-		margin: 1rem 0 0.5rem 0;
-	}
-
-	.markdown-content :global(h4),
-	.markdown-content :global(h5),
-	.markdown-content :global(h6) {
-		font-size: 1rem;
-		font-weight: 600;
-		margin: 0.75rem 0 0.25rem 0;
-	}
-
-	/* Bold and italic */
-	.markdown-content :global(strong) {
-		font-weight: 600;
-	}
-
-	.markdown-content :global(em) {
-		font-style: italic;
-	}
-
-	/* Inline code */
-	.markdown-content :global(.inline-code) {
-		background: rgba(255, 215, 0, 0.1);
-		color: var(--exo-yellow, #ffd700);
-		padding: 0.125rem 0.375rem;
-		border-radius: 0.25rem;
-		font-family: ui-monospace, SFMono-Regular, 'SF Mono', Monaco, Consolas, monospace;
-		font-size: 0.875em;
-	}
-
-	/* Links */
-	.markdown-content :global(a) {
-		color: var(--exo-yellow, #ffd700);
-		text-decoration: underline;
-		text-underline-offset: 2px;
-	}
-
-	.markdown-content :global(a:hover) {
-		opacity: 0.8;
-	}
-
-	/* Lists */
-	.markdown-content :global(ul) {
-		list-style-type: disc;
-		margin-left: 1.5rem;
-		margin-bottom: 1rem;
-	}
-
-	.markdown-content :global(ol) {
-		list-style-type: decimal;
-		margin-left: 1.5rem;
-		margin-bottom: 1rem;
-	}
-
-	.markdown-content :global(li) {
-		margin-bottom: 0.25rem;
-	}
-
-	.markdown-content :global(li::marker) {
-		color: var(--exo-light-gray, #9ca3af);
-	}
-
-	/* Blockquotes */
-	.markdown-content :global(blockquote) {
-		border-left: 3px solid var(--exo-yellow, #ffd700);
-		padding: 0.5rem 1rem;
-		margin: 1rem 0;
-		background: rgba(255, 215, 0, 0.05);
-		border-radius: 0 0.25rem 0.25rem 0;
-	}
-
-	/* Tables */
-	.markdown-content :global(table) {
-		width: 100%;
-		margin: 1rem 0;
-		border-collapse: collapse;
-		font-size: 0.875rem;
-	}
-
-	.markdown-content :global(th) {
-		background: rgba(255, 215, 0, 0.1);
-		border: 1px solid rgba(255, 215, 0, 0.2);
-		padding: 0.5rem;
-		text-align: left;
-		font-weight: 600;
-	}
-
-	.markdown-content :global(td) {
-		border: 1px solid rgba(255, 255, 255, 0.1);
-		padding: 0.5rem;
-	}
-
-	/* Horizontal rule */
-	.markdown-content :global(hr) {
-		border: none;
-		border-top: 1px solid rgba(255, 255, 255, 0.1);
-		margin: 1.5rem 0;
-	}
-
-	/* Code block wrapper */
-	.markdown-content :global(.code-block-wrapper) {
-		margin: 1rem 0;
-		border-radius: 0.5rem;
-		overflow: hidden;
-		border: 1px solid rgba(255, 215, 0, 0.2);
-		background: rgba(0, 0, 0, 0.4);
-	}
-
-	.markdown-content :global(.code-block-header) {
-		display: flex;
-		justify-content: space-between;
-		align-items: center;
-		padding: 0.5rem 0.75rem;
-		background: rgba(255, 215, 0, 0.05);
-		border-bottom: 1px solid rgba(255, 215, 0, 0.1);
-	}
-
-	.markdown-content :global(.code-language) {
-		color: var(--exo-yellow, #ffd700);
-		font-size: 0.7rem;
-		font-weight: 500;
-		text-transform: uppercase;
-		letter-spacing: 0.1em;
-		font-family: ui-monospace, SFMono-Regular, 'SF Mono', Monaco, Consolas, monospace;
-	}
-
-	.markdown-content :global(.copy-code-btn) {
-		display: flex;
-		align-items: center;
-		justify-content: center;
-		padding: 0.25rem;
-		background: transparent;
-		border: none;
-		color: var(--exo-light-gray, #9ca3af);
-		cursor: pointer;
-		transition: color 0.2s;
-		border-radius: 0.25rem;
-	}
-
-	.markdown-content :global(.copy-code-btn:hover) {
-		color: var(--exo-yellow, #ffd700);
-	}
-
-	.markdown-content :global(.copy-code-btn.copied) {
-		color: #22c55e;
-	}
-
-	.markdown-content :global(.code-block-wrapper pre) {
-		margin: 0;
-		padding: 1rem;
-		overflow-x: auto;
-		background: transparent;
-	}
-
-	.markdown-content :global(.code-block-wrapper code) {
-		font-family: ui-monospace, SFMono-Regular, 'SF Mono', Monaco, Consolas, monospace;
-		font-size: 0.8125rem;
-		line-height: 1.5;
-		background: transparent;
-	}
-
-	/* Syntax highlighting - dark theme matching EXO style */
-	.markdown-content :global(.hljs) {
-		color: #e5e7eb;
-	}
-
-	.markdown-content :global(.hljs-keyword),
-	.markdown-content :global(.hljs-selector-tag),
-	.markdown-content :global(.hljs-literal),
-	.markdown-content :global(.hljs-section),
-	.markdown-content :global(.hljs-link) {
-		color: #c084fc;
-	}
-
-	.markdown-content :global(.hljs-string),
-	.markdown-content :global(.hljs-title),
-	.markdown-content :global(.hljs-name),
-	.markdown-content :global(.hljs-type),
-	.markdown-content :global(.hljs-attribute),
-	.markdown-content :global(.hljs-symbol),
-	.markdown-content :global(.hljs-bullet),
-	.markdown-content :global(.hljs-addition),
-	.markdown-content :global(.hljs-variable),
-	.markdown-content :global(.hljs-template-tag),
-	.markdown-content :global(.hljs-template-variable) {
-		color: #fbbf24;
-	}
-
-	.markdown-content :global(.hljs-comment),
-	.markdown-content :global(.hljs-quote),
-	.markdown-content :global(.hljs-deletion),
-	.markdown-content :global(.hljs-meta) {
-		color: #6b7280;
-	}
-
-	.markdown-content :global(.hljs-number),
-	.markdown-content :global(.hljs-regexp),
-	.markdown-content :global(.hljs-literal),
-	.markdown-content :global(.hljs-built_in) {
-		color: #34d399;
-	}
-
-	.markdown-content :global(.hljs-function),
-	.markdown-content :global(.hljs-class .hljs-title) {
-		color: #60a5fa;
-	}
-
-	/* KaTeX math styling */
-	.markdown-content :global(.katex) {
-		font-size: 1.1em;
-	}
-
-	.markdown-content :global(.katex-display) {
-		margin: 1rem 0;
-		overflow-x: auto;
-		overflow-y: hidden;
-		padding: 0.5rem 0;
-	}
-
-	.markdown-content :global(.katex-display > .katex) {
-		text-align: center;
-	}
-
-	.markdown-content :global(.math-error) {
-		color: #f87171;
-		font-family: ui-monospace, SFMono-Regular, 'SF Mono', Monaco, Consolas, monospace;
-		font-size: 0.875em;
-		background: rgba(248, 113, 113, 0.1);
-		padding: 0.125rem 0.25rem;
-		border-radius: 0.25rem;
-	}
-</style>

dashboard/src/lib/components/ModelCard.svelte

@@ -1,6 +1,5 @@
 <script lang="ts">
-	import type { DownloadProgress, NodeInfo, PlacementPreview, TopologyEdge } from '$lib/stores/app.svelte';
-	import { debugMode, topologyData } from '$lib/stores/app.svelte';
+	import type { DownloadProgress, NodeInfo, PlacementPreview } from '$lib/stores/app.svelte';
 
 interface Props {
 		model: { id: string; name?: string; storage_size_megabytes?: number };
@@ -207,8 +206,12 @@ function toggleNodeDetails(nodeId: string): void {
 		const centerY = topoHeight / 2;
 		const radius = numNodes === 1 ? 0 : numNodes === 2 ? 45 : Math.min(topoWidth, topoHeight) * 0.32;
 		
-		// Only use API preview data - no local estimation
+		// Use API preview data if available
 		const hasApiPreview = apiPreview !== null && apiPreview.error === null && apiPreview.memory_delta_by_node !== null;
+		const canFit = hasApiPreview ? true : (() => {
+			const totalAvailable = nodeArray.reduce((sum, n) => sum + n.availableGB, 0);
+			return totalAvailable >= estimatedMemory;
+		})();
 		const error = apiPreview?.error ?? null;
 		
 		let placementNodes: Array<{ 
@@ -229,140 +232,135 @@ function toggleNodeDetails(nodeId: string): void {
 			modelFillHeight: number;
 		}> = [];
 		
-		// Use API placement data directly
-		const memoryDelta = apiPreview?.memory_delta_by_node ?? {};
-		placementNodes = nodeArray.map((n, i) => {
-			const deltaBytes = memoryDelta[n.id] ?? 0;
-			const modelUsageGB = deltaBytes / (1024 * 1024 * 1024);
-			const isUsed = deltaBytes > 0;
-			const angle = numNodes === 1 ? 0 : (i / numNodes) * Math.PI * 2 - Math.PI / 2;
-			const safeTotal = Math.max(n.totalGB, 0.001);
-			const currentPercent = clampPercent((n.usedGB / safeTotal) * 100);
-			const newPercent = clampPercent(((n.usedGB + modelUsageGB) / safeTotal) * 100);
-			const screenHeight = iconSize * 0.58;
+		if (hasApiPreview && apiPreview.memory_delta_by_node) {
+			// Use API placement data
+			const memoryDelta = apiPreview.memory_delta_by_node;
+			placementNodes = nodeArray.map((n, i) => {
+				const deltaBytes = memoryDelta[n.id] ?? 0;
+				const modelUsageGB = deltaBytes / (1024 * 1024 * 1024);
+				const isUsed = deltaBytes > 0;
+				const angle = numNodes === 1 ? 0 : (i / numNodes) * Math.PI * 2 - Math.PI / 2;
+				const safeTotal = Math.max(n.totalGB, 0.001);
+				const currentPercent = clampPercent((n.usedGB / safeTotal) * 100);
+				const newPercent = clampPercent(((n.usedGB + modelUsageGB) / safeTotal) * 100);
+				const screenHeight = iconSize * 0.58;
+				
+				return {
+					id: n.id,
+					deviceName: n.deviceName,
+					deviceType: n.deviceType,
+					totalGB: n.totalGB,
+					currentUsedGB: n.usedGB,
+					modelUsageGB,
+					currentPercent,
+					newPercent,
+					isUsed,
+					x: centerX + Math.cos(angle) * radius,
+					y: centerY + Math.sin(angle) * radius,
+					iconSize,
+					screenHeight,
+					currentFillHeight: screenHeight * (currentPercent / 100),
+					modelFillHeight: screenHeight * ((newPercent - currentPercent) / 100)
+				};
+			});
+		} else if (apiPreview?.error) {
+			// API returned an error - model can't fit, show all nodes as unused
+			placementNodes = nodeArray.map((n, i) => {
+				const angle = numNodes === 1 ? 0 : (i / numNodes) * Math.PI * 2 - Math.PI / 2;
+				const safeTotal = Math.max(n.totalGB, 0.001);
+				const currentPercent = clampPercent((n.usedGB / safeTotal) * 100);
+				const screenHeight = iconSize * 0.58;
+				
+				return {
+					id: n.id,
+					deviceName: n.deviceName,
+					deviceType: n.deviceType,
+					totalGB: n.totalGB,
+					currentUsedGB: n.usedGB,
+					modelUsageGB: 0,
+					currentPercent,
+					newPercent: currentPercent,
+					isUsed: false,
+					x: centerX + Math.cos(angle) * radius,
+					y: centerY + Math.sin(angle) * radius,
+					iconSize,
+					screenHeight,
+					currentFillHeight: screenHeight * (currentPercent / 100),
+					modelFillHeight: 0
+				};
+			});
+		} else {
+			// Fallback: local estimation based on sharding strategy
+			const memoryNeeded = estimatedMemory;
 			
-			return {
-				id: n.id,
-				deviceName: n.deviceName,
-				deviceType: n.deviceType,
-				totalGB: n.totalGB,
-				currentUsedGB: n.usedGB,
-				modelUsageGB,
-				currentPercent,
-				newPercent,
-				isUsed,
-				x: centerX + Math.cos(angle) * radius,
-				y: centerY + Math.sin(angle) * radius,
-				iconSize,
-				screenHeight,
-				currentFillHeight: screenHeight * (currentPercent / 100),
-				modelFillHeight: screenHeight * ((newPercent - currentPercent) / 100)
-			};
-		});
+			if (sharding === 'Pipeline') {
+				const memoryPerNode = memoryNeeded / numNodes;
+				placementNodes = nodeArray.map((n, i) => {
+					const angle = numNodes === 1 ? 0 : (i / numNodes) * Math.PI * 2 - Math.PI / 2;
+					const safeTotal = Math.max(n.totalGB, 0.001);
+					const currentPercent = clampPercent((n.usedGB / safeTotal) * 100);
+					const newPercent = clampPercent(((n.usedGB + memoryPerNode) / safeTotal) * 100);
+					const screenHeight = iconSize * 0.58;
+					
+					return {
+						id: n.id,
+						deviceName: n.deviceName,
+						deviceType: n.deviceType,
+						totalGB: n.totalGB,
+						currentUsedGB: n.usedGB,
+						modelUsageGB: memoryPerNode,
+						currentPercent,
+						newPercent,
+						isUsed: true,
+						x: centerX + Math.cos(angle) * radius,
+						y: centerY + Math.sin(angle) * radius,
+						iconSize,
+						screenHeight,
+						currentFillHeight: screenHeight * (currentPercent / 100),
+						modelFillHeight: screenHeight * ((newPercent - currentPercent) / 100)
+					};
+				});
+			} else {
+				let remaining = memoryNeeded;
+				placementNodes = nodeArray.map((n, i) => {
+					const allocated = Math.min(remaining, n.availableGB);
+					remaining -= allocated;
+					const isUsed = allocated > 0;
+					const angle = numNodes === 1 ? 0 : (i / numNodes) * Math.PI * 2 - Math.PI / 2;
+					const safeTotal = Math.max(n.totalGB, 0.001);
+					const currentPercent = clampPercent((n.usedGB / safeTotal) * 100);
+					const newPercent = clampPercent(((n.usedGB + allocated) / safeTotal) * 100);
+					const screenHeight = iconSize * 0.58;
+					
+					return {
+						id: n.id,
+						deviceName: n.deviceName,
+						deviceType: n.deviceType,
+						totalGB: n.totalGB,
+						currentUsedGB: n.usedGB,
+						modelUsageGB: allocated,
+						currentPercent,
+						newPercent,
+						isUsed,
+						x: centerX + Math.cos(angle) * radius,
+						y: centerY + Math.sin(angle) * radius,
+						iconSize,
+						screenHeight,
+						currentFillHeight: screenHeight * (currentPercent / 100),
+						modelFillHeight: screenHeight * ((newPercent - currentPercent) / 100)
+					};
+				});
+			}
+		}
 		
 		const totalAvailable = nodeArray.reduce((sum, n) => sum + n.availableGB, 0);
-		return { nodes: placementNodes, canFit: hasApiPreview, totalAvailable, topoWidth, topoHeight, error };
+		return { nodes: placementNodes, canFit: hasApiPreview || canFit, totalAvailable, topoWidth, topoHeight, error };
 	});
 	
 	const canFit = $derived(apiPreview ? apiPreview.error === null : placementPreview().canFit);
 	const placementError = $derived(apiPreview?.error ?? null);
 	const nodeCount = $derived(nodeList().length);
 	const filterId = $derived(model.id.replace(/[^a-zA-Z0-9]/g, ''));
-	
-	// Debug mode state
-	const isDebugMode = $derived(debugMode());
-	const topology = $derived(topologyData());
-	const isRdma = $derived(runtime === 'MlxIbv' || runtime === 'MlxJaccl');
-	
-	// Get interface name for an IP from node data
-	function getInterfaceForIp(nodeId: string, ip?: string): string | null {
-		if (!ip || !topology?.nodes) return null;
-		
-		// Strip port if present
-		const cleanIp = ip.includes(':') && !ip.includes('[') ? ip.split(':')[0] : ip;
-		
-		// Check specified node first
-		const node = topology.nodes[nodeId];
-		if (node) {
-			const match = node.network_interfaces?.find((iface) =>
-				(iface.addresses || []).some((addr) => addr === cleanIp || addr === ip)
-			);
-			if (match?.name) return match.name;
-			
-			const mapped = node.ip_to_interface?.[cleanIp] || node.ip_to_interface?.[ip];
-			if (mapped) return mapped;
-		}
-		
-		// Fallback: check all nodes
-		for (const [, otherNode] of Object.entries(topology.nodes)) {
-			if (!otherNode) continue;
-			const match = otherNode.network_interfaces?.find((iface) =>
-				(iface.addresses || []).some((addr) => addr === cleanIp || addr === ip)
-			);
-			if (match?.name) return match.name;
-			
-			const mapped = otherNode.ip_to_interface?.[cleanIp] || otherNode.ip_to_interface?.[ip];
-			if (mapped) return mapped;
-		}
-		
-		return null;
-	}
-	
-	// Get directional arrow based on node positions
-	function getArrow(fromNode: { x: number; y: number }, toNode: { x: number; y: number }): string {
-		const dx = toNode.x - fromNode.x;
-		const dy = toNode.y - fromNode.y;
-		const absX = Math.abs(dx);
-		const absY = Math.abs(dy);
-		
-		if (absX > absY * 2) {
-			return dx > 0 ? '→' : '←';
-		} else if (absY > absX * 2) {
-			return dy > 0 ? '↓' : '↑';
-		} else {
-			if (dx > 0 && dy > 0) return '↘';
-			if (dx > 0 && dy < 0) return '↗';
-			if (dx < 0 && dy > 0) return '↙';
-			return '↖';
-		}
-	}
-
-	// Get connection info for edges between two nodes
-	// Returns exactly one connection per direction (A→B and B→A), preferring non-loopback
-	function getConnectionInfo(nodeId1: string, nodeId2: string): Array<{ ip: string; iface: string | null; from: string; to: string }> {
-		if (!topology?.edges) return [];
-		
-		// Collect candidates for each direction
-		const aToBCandidates: Array<{ ip: string; iface: string | null }> = [];
-		const bToACandidates: Array<{ ip: string; iface: string | null }> = [];
-		
-		for (const edge of topology.edges) {
-			const ip = edge.sendBackIp || '?';
-			const iface = edge.sendBackInterface || getInterfaceForIp(edge.source, ip);
-			
-			if (edge.source === nodeId1 && edge.target === nodeId2) {
-				aToBCandidates.push({ ip, iface });
-			} else if (edge.source === nodeId2 && edge.target === nodeId1) {
-				bToACandidates.push({ ip, iface });
-			}
-		}
-		
-		// Pick best (prefer non-loopback)
-		const pickBest = (candidates: Array<{ ip: string; iface: string | null }>) => {
-			if (candidates.length === 0) return null;
-			return candidates.find(c => !c.ip.startsWith('127.')) || candidates[0];
-		};
-		
-		const result: Array<{ ip: string; iface: string | null; from: string; to: string }> = [];
-		
-		const bestAtoB = pickBest(aToBCandidates);
-		if (bestAtoB) result.push({ ...bestAtoB, from: nodeId1, to: nodeId2 });
-		
-		const bestBtoA = pickBest(bToACandidates);
-		if (bestBtoA) result.push({ ...bestBtoA, from: nodeId2, to: nodeId1 });
-		
-		return result;
-	}
 </script>
 
 <div class="relative group">
@@ -455,26 +453,6 @@ function toggleNodeDetails(nodeId: string): void {
 					
 					<!-- Connection lines between nodes (if multiple) -->
 					{#if preview.nodes.length > 1}
-						{@const usedNodes = preview.nodes.filter(n => n.isUsed)}
-						{@const nodePositions = Object.fromEntries(preview.nodes.map(n => [n.id, { x: n.x, y: n.y }]))}
-						{@const allConnections = isDebugMode && usedNodes.length > 1 ? (() => {
-							const conns: Array<{ ip: string; iface: string | null; from: string; to: string; midX: number; midY: number; arrow: string }> = [];
-							for (let i = 0; i < usedNodes.length; i++) {
-								for (let j = i + 1; j < usedNodes.length; j++) {
-									const n1 = usedNodes[i];
-									const n2 = usedNodes[j];
-									const midX = (n1.x + n2.x) / 2;
-									const midY = (n1.y + n2.y) / 2;
-									for (const c of getConnectionInfo(n1.id, n2.id)) {
-										const fromPos = nodePositions[c.from];
-										const toPos = nodePositions[c.to];
-										const arrow = fromPos && toPos ? getArrow(fromPos, toPos) : '→';
-										conns.push({ ...c, midX, midY, arrow });
-									}
-								}
-							}
-							return conns;
-						})() : []}
 						{#each preview.nodes as node, i}
 							{#each preview.nodes.slice(i + 1) as node2}
 								<line 
@@ -486,43 +464,6 @@ function toggleNodeDetails(nodeId: string): void {
 								/>
 							{/each}
 						{/each}
-						<!-- Debug: Show connection IPs/interfaces in corners -->
-						{#if isDebugMode && allConnections.length > 0}
-							{@const centerX = preview.topoWidth / 2}
-							{@const centerY = preview.topoHeight / 2}
-							{@const quadrants = {
-								topLeft: allConnections.filter(c => c.midX < centerX && c.midY < centerY),
-								topRight: allConnections.filter(c => c.midX >= centerX && c.midY < centerY),
-								bottomLeft: allConnections.filter(c => c.midX < centerX && c.midY >= centerY),
-								bottomRight: allConnections.filter(c => c.midX >= centerX && c.midY >= centerY)
-							}}
-							{@const padding = 4}
-							{@const lineHeight = 8}
-							<!-- Top Left -->
-							{#each quadrants.topLeft as conn, idx}
-								<text x={padding} y={padding + idx * lineHeight} text-anchor="start" dominant-baseline="hanging" font-size="6" font-family="SF Mono, Monaco, monospace" fill={conn.iface ? 'rgba(255,255,255,0.85)' : 'rgba(248,113,113,0.85)'}>
-									{conn.arrow} {isRdma ? (conn.iface || '?') : `${conn.ip}${conn.iface ? ` (${conn.iface})` : ''}`}
-								</text>
-							{/each}
-							<!-- Top Right -->
-							{#each quadrants.topRight as conn, idx}
-								<text x={preview.topoWidth - padding} y={padding + idx * lineHeight} text-anchor="end" dominant-baseline="hanging" font-size="6" font-family="SF Mono, Monaco, monospace" fill={conn.iface ? 'rgba(255,255,255,0.85)' : 'rgba(248,113,113,0.85)'}>
-									{conn.arrow} {isRdma ? (conn.iface || '?') : `${conn.ip}${conn.iface ? ` (${conn.iface})` : ''}`}
-								</text>
-							{/each}
-							<!-- Bottom Left -->
-							{#each quadrants.bottomLeft as conn, idx}
-								<text x={padding} y={preview.topoHeight - padding - (quadrants.bottomLeft.length - 1 - idx) * lineHeight} text-anchor="start" dominant-baseline="auto" font-size="6" font-family="SF Mono, Monaco, monospace" fill={conn.iface ? 'rgba(255,255,255,0.85)' : 'rgba(248,113,113,0.85)'}>
-									{conn.arrow} {isRdma ? (conn.iface || '?') : `${conn.ip}${conn.iface ? ` (${conn.iface})` : ''}`}
-								</text>
-							{/each}
-							<!-- Bottom Right -->
-							{#each quadrants.bottomRight as conn, idx}
-								<text x={preview.topoWidth - padding} y={preview.topoHeight - padding - (quadrants.bottomRight.length - 1 - idx) * lineHeight} text-anchor="end" dominant-baseline="auto" font-size="6" font-family="SF Mono, Monaco, monospace" fill={conn.iface ? 'rgba(255,255,255,0.85)' : 'rgba(248,113,113,0.85)'}>
-									{conn.arrow} {isRdma ? (conn.iface || '?') : `${conn.ip}${conn.iface ? ` (${conn.iface})` : ''}`}
-								</text>
-							{/each}
-						{/if}
 					{/if}
 					
 					{#each preview.nodes as node}

dashboard/src/lib/components/TopologyGraph.svelte

@@ -24,36 +24,19 @@ function getNodeLabel(nodeId: string): string {
 
 function getInterfaceLabel(nodeId: string, ip?: string): { label: string; missing: boolean } {
 	if (!ip) return { label: '?', missing: true };
-	
-	// Strip port if present (e.g., "192.168.1.1:8080" -> "192.168.1.1")
-	const cleanIp = ip.includes(':') && !ip.includes('[') ? ip.split(':')[0] : ip;
-	
-	// Helper to check a node's interfaces
-	function checkNode(node: typeof data.nodes[string]): string | null {
-		if (!node) return null;
-		
-		const matchFromInterfaces = node.network_interfaces?.find((iface) =>
-			(iface.addresses || []).some((addr) => addr === cleanIp || addr === ip)
-		);
-		if (matchFromInterfaces?.name) {
-			return matchFromInterfaces.name;
-		}
+	const node = data?.nodes?.[nodeId];
+	if (!node) return { label: '?', missing: true };
 
-		const mapped = node.ip_to_interface?.[cleanIp] || node.ip_to_interface?.[ip];
-		if (mapped && mapped.trim().length > 0) {
-			return mapped;
-		}
-		return null;
+	const matchFromInterfaces = node.network_interfaces?.find((iface) =>
+		(iface.addresses || []).some((addr) => addr === ip)
+	);
+	if (matchFromInterfaces?.name) {
+		return { label: matchFromInterfaces.name, missing: false };
 	}
-	
-	// Try specified node first
-	const result = checkNode(data?.nodes?.[nodeId]);
-	if (result) return { label: result, missing: false };
-	
-	// Fallback: search all nodes for this IP
-	for (const [, otherNode] of Object.entries(data?.nodes || {})) {
-		const otherResult = checkNode(otherNode);
-		if (otherResult) return { label: otherResult, missing: false };
+
+	const mapped = node.ip_to_interface?.[ip];
+	if (mapped && mapped.trim().length > 0) {
+		return { label: mapped, missing: false };
 	}
 
 	return { label: '?', missing: true };
@@ -84,7 +67,6 @@ function wrapLine(text: string, maxLen: number): string[] {
 	return lines;
 }
 
-
 	// Apple logo path for MacBook Pro screen
 	const APPLE_LOGO_PATH = "M788.1 340.9c-5.8 4.5-108.2 62.2-108.2 190.5 0 148.4 130.3 200.9 134.2 202.2-.6 3.2-20.7 71.9-68.7 141.9-42.8 61.6-87.5 123.1-155.5 123.1s-85.5-39.5-164-39.5c-76.5 0-103.7 40.8-165.9 40.8s-105.6-57-155.5-127C46.7 790.7 0 663 0 541.8c0-194.4 126.4-297.5 250.8-297.5 66.1 0 121.2 43.4 162.7 43.4 39.5 0 101.1-46 176.3-46 28.5 0 130.9 2.6 198.3 99.2zm-234-181.5c31.1-36.9 53.1-88.1 53.1-139.3 0-7.1-.6-14.3-1.9-20.1-50.6 1.9-110.8 33.7-147.1 75.8-28.5 32.4-55.1 83.6-55.1 135.5 0 7.8 1.3 15.6 1.9 18.1 3.2.6 8.4 1.3 13.6 1.3 45.4 0 102.5-30.4 135.5-71.3z";
 	const LOGO_NATIVE_WIDTH = 814;
@@ -256,7 +238,6 @@ function wrapLine(text: string, maxLen: number): string[] {
 		const debugLabelsGroup = svg.append('g').attr('class', 'debug-edge-labels');
 
 		const pairMap = new Map<string, { a: string; b: string; aToB: boolean; bToA: boolean; connections: Array<{ from: string; to: string; ip: string; ifaceLabel: string; missingIface: boolean }> }>();
-		let debugEdgeLabels: Array<{ connections: typeof pairMap extends Map<string, infer V> ? V['connections'] : never; isLeft: boolean; isTop: boolean; mx: number; my: number }> | null = null;
 		edges.forEach(edge => {
 			if (!edge.source || !edge.target || edge.source === edge.target) return;
 			if (!positionById[edge.source] || !positionById[edge.target]) return;
@@ -333,97 +314,109 @@ function wrapLine(text: string, maxLen: number): string[] {
 					.attr('marker-end', 'url(#arrowhead)');
 			}
 
-			// Collect debug labels for later positioning at edges
 			if (debugEnabled && entry.connections.length > 0) {
-				// Determine which side of viewport based on edge midpoint
-				const isLeft = mx < centerX;
-				const isTop = my < safeCenterY;
-				
-				// Store for batch rendering after all edges processed
-				if (!debugEdgeLabels) debugEdgeLabels = [];
-				debugEdgeLabels.push({
-					connections: entry.connections,
-					isLeft,
-					isTop,
-					mx,
-					my
-				});
-			}
-		});
+				const maxBoxes = 6;
+				const fontSize = isMinimized ? 8 : 9;
+				const lineGap = 2;
+				const labelOffsetOut = Math.max(140, minDimension * 0.38);
+				const labelOffsetSide = isMinimized ? 16 : 20;
+				const boxWidth = 170;
+				const maxLineLen = 26;
 
-		// Render debug labels at viewport edges/corners
-		if (debugEdgeLabels && debugEdgeLabels.length > 0) {
-			const fontSize = isMinimized ? 10 : 12;
-			const lineHeight = fontSize + 4;
-			const padding = 10;
-			
-			// Helper to get arrow based on direction vector
-			function getArrow(fromId: string, toId: string): string {
-				const fromPos = positionById[fromId];
-				const toPos = positionById[toId];
-				if (!fromPos || !toPos) return '→';
-				
-				const dirX = toPos.x - fromPos.x;
-				const dirY = toPos.y - fromPos.y;
-				const absX = Math.abs(dirX);
-				const absY = Math.abs(dirY);
-				
-				if (absX > absY * 2) {
-					return dirX > 0 ? '→' : '←';
-				} else if (absY > absX * 2) {
-					return dirY > 0 ? '↓' : '↑';
-				} else {
-					if (dirX > 0 && dirY > 0) return '↘';
-					if (dirX > 0 && dirY < 0) return '↗';
-					if (dirX < 0 && dirY > 0) return '↙';
-					return '↖';
+				const connections = entry.connections.slice(0, maxBoxes);
+				if (entry.connections.length > maxBoxes) {
+					const remaining = entry.connections.length - maxBoxes;
+					connections.push({
+						from: '',
+						to: '',
+						ip: `(+${remaining} more)`,
+						ifaceLabel: '',
+						missingIface: false
+					});
 				}
-			}
-			
-			// Group by quadrant: topLeft, topRight, bottomLeft, bottomRight
-			const quadrants: Record<string, typeof debugEdgeLabels> = {
-				topLeft: [],
-				topRight: [],
-				bottomLeft: [],
-				bottomRight: []
-			};
-			
-			debugEdgeLabels.forEach(edge => {
-				const key = (edge.isTop ? 'top' : 'bottom') + (edge.isLeft ? 'Left' : 'Right');
-				quadrants[key].push(edge);
-			});
-			
-			// Render each quadrant
-			Object.entries(quadrants).forEach(([quadrant, edges]) => {
-				if (edges.length === 0) return;
-				
-				const isLeft = quadrant.includes('Left');
-				const isTop = quadrant.includes('top');
-				
-				let baseX = isLeft ? padding : width - padding;
-				let baseY = isTop ? padding : height - padding;
-				const textAnchor = isLeft ? 'start' : 'end';
-				
-				let currentY = baseY;
-				
-				edges.forEach(edge => {
-					edge.connections.forEach(conn => {
-						const arrow = getArrow(conn.from, conn.to);
-						const label = `${arrow} ${conn.ip} ${conn.ifaceLabel}`;
-						debugLabelsGroup.append('text')
-							.attr('x', baseX)
-							.attr('y', currentY)
-							.attr('text-anchor', textAnchor)
-							.attr('dominant-baseline', isTop ? 'hanging' : 'auto')
+
+				let dirX = mx - centerX;
+				let dirY = my - centerY;
+				const dirLen = Math.hypot(dirX, dirY);
+				if (dirLen < 1) {
+					dirX = -uy;
+					dirY = ux;
+				} else {
+					dirX /= dirLen;
+					dirY /= dirLen;
+				}
+
+				const nx = -dirY;
+				const ny = dirX;
+
+				const labelXRaw = mx + dirX * labelOffsetOut + nx * labelOffsetSide;
+				const labelYRaw = my + dirY * labelOffsetOut + ny * labelOffsetSide;
+				const clampPad = Math.min(120, minDimension * 0.12);
+				const labelX = Math.max(clampPad, Math.min(width - clampPad, labelXRaw));
+				const labelY = Math.max(clampPad, Math.min(height - clampPad, labelYRaw));
+
+				const labelGroup = debugLabelsGroup.append('g')
+					.attr('transform', `translate(${labelX}, ${labelY})`);
+
+				const textGroup = labelGroup.append('g');
+
+				connections.forEach((conn, idx) => {
+					const rawLines = conn.from && conn.to
+						? [
+							`${getNodeLabel(conn.from)}→${getNodeLabel(conn.to)}`,
+							`${conn.ip}`,
+							`${conn.ifaceLabel}`
+						]
+						: [conn.ip];
+
+					const wrapped = rawLines.flatMap(line => wrapLine(line, maxLineLen));
+
+					wrapped.forEach((line, lineIdx) => {
+						textGroup.append('text')
+							.attr('x', 0)
+							.attr('y', (idx * (wrapped.length * (fontSize + lineGap))) + lineIdx * (fontSize + lineGap))
+							.attr('text-anchor', 'middle')
+							.attr('dominant-baseline', 'hanging')
 							.attr('font-size', fontSize)
 							.attr('font-family', 'SF Mono, monospace')
-							.attr('fill', conn.missingIface ? 'rgba(248,113,113,0.9)' : 'rgba(255,255,255,0.85)')
-							.text(label);
-						currentY += isTop ? lineHeight : -lineHeight;
+							.attr('fill', conn.missingIface ? 'rgba(248,113,113,0.9)' : 'rgba(255,255,255,0.9)')
+							.text(line);
 					});
 				});
-			});
-		}
+
+				const bbox = textGroup.node()?.getBBox();
+				if (bbox) {
+					const paddedWidth = Math.max(boxWidth, bbox.width + 14);
+					const boxHeight = bbox.height + 8;
+					const boxMinX = labelX - paddedWidth / 2;
+					const boxMaxX = labelX + paddedWidth / 2;
+					const boxMinY = labelY + bbox.y - 4;
+					const boxMaxY = boxMinY + boxHeight;
+
+					const clampPadDynamic = Math.min(140, minDimension * 0.18);
+					let shiftX = 0;
+					let shiftY = 0;
+					if (boxMinX < clampPadDynamic) shiftX = clampPadDynamic - boxMinX;
+					if (boxMaxX > width - clampPadDynamic) shiftX = (width - clampPadDynamic) - boxMaxX;
+					if (boxMinY < clampPadDynamic) shiftY = clampPadDynamic - boxMinY;
+					if (boxMaxY > height - clampPadDynamic) shiftY = (height - clampPadDynamic) - boxMaxY;
+
+					const finalX = labelX + shiftX;
+					const finalY = labelY + shiftY;
+					labelGroup.attr('transform', `translate(${finalX}, ${finalY})`);
+
+					labelGroup.insert('rect', 'g')
+						.attr('x', -paddedWidth / 2)
+						.attr('y', bbox.y - 4)
+						.attr('width', paddedWidth)
+						.attr('height', boxHeight)
+						.attr('rx', 4)
+						.attr('fill', 'rgba(0,0,0,0.75)')
+						.attr('stroke', 'rgba(255,255,255,0.12)')
+						.attr('stroke-width', 0.6);
+				}
+			}
+		});
 
 		// Draw nodes
 		const nodesGroup = svg.append('g').attr('class', 'nodes-group');
@@ -975,5 +968,4 @@ function wrapLine(text: string, maxLen: number): string[] {
 		from { stroke-dashoffset: 0; }
 		to { stroke-dashoffset: -10; }
 	}
-
 </style>

dashboard/src/lib/components/index.ts

@@ -4,5 +4,4 @@ export { default as ChatMessages } from './ChatMessages.svelte';
 export { default as ChatAttachments } from './ChatAttachments.svelte';
 export { default as ChatSidebar } from './ChatSidebar.svelte';
 export { default as ModelCard } from './ModelCard.svelte';
-export { default as MarkdownContent } from './MarkdownContent.svelte';
 

dashboard/src/lib/stores/app.svelte.ts

@@ -327,8 +327,6 @@ class AppStore {
 	isTopologyMinimized = $state(false);
 	isSidebarOpen = $state(false); // Hidden by default, shown when in chat mode
 	debugMode = $state(false);
-	topologyOnlyMode = $state(false);
-	chatSidebarVisible = $state(true); // Shown by default
 	
 	private fetchInterval: ReturnType<typeof setInterval> | null = null;
 	private previewsInterval: ReturnType<typeof setInterval> | null = null;
@@ -339,8 +337,6 @@ class AppStore {
 			this.startPolling();
 			this.loadConversationsFromStorage();
 			this.loadDebugModeFromStorage();
-			this.loadTopologyOnlyModeFromStorage();
-			this.loadChatSidebarVisibleFromStorage();
 		}
 	}
 
@@ -398,44 +394,6 @@ class AppStore {
 		}
 	}
 
-	private loadTopologyOnlyModeFromStorage() {
-		try {
-			const stored = localStorage.getItem('exo-topology-only-mode');
-			if (stored !== null) {
-				this.topologyOnlyMode = stored === 'true';
-			}
-		} catch (error) {
-			console.error('Failed to load topology only mode:', error);
-		}
-	}
-
-	private saveTopologyOnlyModeToStorage() {
-		try {
-			localStorage.setItem('exo-topology-only-mode', this.topologyOnlyMode ? 'true' : 'false');
-		} catch (error) {
-			console.error('Failed to save topology only mode:', error);
-		}
-	}
-
-	private loadChatSidebarVisibleFromStorage() {
-		try {
-			const stored = localStorage.getItem('exo-chat-sidebar-visible');
-			if (stored !== null) {
-				this.chatSidebarVisible = stored === 'true';
-			}
-		} catch (error) {
-			console.error('Failed to load chat sidebar visibility:', error);
-		}
-	}
-
-	private saveChatSidebarVisibleToStorage() {
-		try {
-			localStorage.setItem('exo-chat-sidebar-visible', this.chatSidebarVisible ? 'true' : 'false');
-		} catch (error) {
-			console.error('Failed to save chat sidebar visibility:', error);
-		}
-	}
-
 	/**
 	 * Create a new conversation
 	 */
@@ -740,34 +698,6 @@ class AppStore {
 		this.saveDebugModeToStorage();
 	}
 
-	getTopologyOnlyMode(): boolean {
-		return this.topologyOnlyMode;
-	}
-
-	setTopologyOnlyMode(enabled: boolean) {
-		this.topologyOnlyMode = enabled;
-		this.saveTopologyOnlyModeToStorage();
-	}
-
-	toggleTopologyOnlyMode() {
-		this.topologyOnlyMode = !this.topologyOnlyMode;
-		this.saveTopologyOnlyModeToStorage();
-	}
-
-	getChatSidebarVisible(): boolean {
-		return this.chatSidebarVisible;
-	}
-
-	setChatSidebarVisible(visible: boolean) {
-		this.chatSidebarVisible = visible;
-		this.saveChatSidebarVisibleToStorage();
-	}
-
-	toggleChatSidebarVisible() {
-		this.chatSidebarVisible = !this.chatSidebarVisible;
-		this.saveChatSidebarVisibleToStorage();
-	}
-
 	startPolling() {
 		this.fetchState();
 		this.fetchInterval = setInterval(() => this.fetchState(), 1000);
@@ -958,6 +888,8 @@ class AppStore {
 		
 		if (lastUserIndex === -1) return;
 		
+		const lastUserMessage = this.messages[lastUserIndex];
+		
 		// Remove any messages after the user message
 		this.messages = this.messages.slice(0, lastUserIndex + 1);
 		
@@ -998,10 +930,7 @@ class AppStore {
 			}
 			
 			if (!modelToUse) {
-				const idx = this.messages.findIndex(m => m.id === assistantMessage.id);
-				if (idx !== -1) {
-					this.messages[idx].content = 'Error: No model available. Please launch an instance first.';
-				}
+				assistantMessage.content = 'Error: No model available. Please launch an instance first.';
 				this.isLoading = false;
 				this.updateActiveConversation();
 				return;
@@ -1019,10 +948,7 @@ class AppStore {
 			
 			if (!response.ok) {
 				const errorText = await response.text();
-				const idx = this.messages.findIndex(m => m.id === assistantMessage.id);
-				if (idx !== -1) {
-					this.messages[idx].content = `Error: ${response.status} - ${errorText}`;
-				}
+				assistantMessage.content = `Error: ${response.status} - ${errorText}`;
 				this.isLoading = false;
 				this.updateActiveConversation();
 				return;
@@ -1030,10 +956,7 @@ class AppStore {
 			
 			const reader = response.body?.getReader();
 			if (!reader) {
-				const idx = this.messages.findIndex(m => m.id === assistantMessage.id);
-				if (idx !== -1) {
-					this.messages[idx].content = 'Error: No response stream available';
-				}
+				assistantMessage.content = 'Error: No response stream available';
 				this.isLoading = false;
 				this.updateActiveConversation();
 				return;
@@ -1061,16 +984,9 @@ class AppStore {
 							const delta = json.choices?.[0]?.delta?.content;
 							if (delta) {
 								fullContent += delta;
-								const { displayContent, thinkingContent } = this.stripThinkingTags(fullContent);
+								const { displayContent } = this.stripThinkingTags(fullContent);
 								this.currentResponse = displayContent;
-								
-								// Update the assistant message in place (triggers Svelte reactivity)
-								const idx = this.messages.findIndex(m => m.id === assistantMessage.id);
-								if (idx !== -1) {
-									this.messages[idx].content = displayContent;
-									this.messages[idx].thinking = thinkingContent || undefined;
-								}
-								this.persistActiveConversation();
+								assistantMessage.content = displayContent;
 							}
 						} catch {
 							// Skip malformed JSON
@@ -1079,25 +995,16 @@ class AppStore {
 				}
 			}
 			
-			// Final cleanup of the message
-			const { displayContent, thinkingContent } = this.stripThinkingTags(fullContent);
-			const idx = this.messages.findIndex(m => m.id === assistantMessage.id);
-			if (idx !== -1) {
-				this.messages[idx].content = displayContent;
-				this.messages[idx].thinking = thinkingContent || undefined;
-			}
-			this.persistActiveConversation();
-			
-		} catch (error) {
-			const idx = this.messages.findIndex(m => m.id === assistantMessage.id);
-			if (idx !== -1) {
-				this.messages[idx].content = `Error: ${error instanceof Error ? error.message : 'Unknown error'}`;
-			}
-			this.persistActiveConversation();
-		} finally {
-			this.isLoading = false;
+			const { displayContent } = this.stripThinkingTags(fullContent);
+			assistantMessage.content = displayContent;
 			this.currentResponse = '';
 			this.updateActiveConversation();
+			
+		} catch (error) {
+			assistantMessage.content = `Error: ${error instanceof Error ? error.message : 'Unknown error'}`;
+			this.updateActiveConversation();
+		} finally {
+			this.isLoading = false;
 		}
 	}
 
@@ -1457,8 +1364,6 @@ export const lastUpdate = () => appStore.lastUpdate;
 export const isTopologyMinimized = () => appStore.isTopologyMinimized;
 export const selectedChatModel = () => appStore.selectedChatModel;
 export const debugMode = () => appStore.getDebugMode();
-export const topologyOnlyMode = () => appStore.getTopologyOnlyMode();
-export const chatSidebarVisible = () => appStore.getChatSidebarVisible();
 
 // Actions
 export const startChat = () => appStore.startChat();
@@ -1486,9 +1391,5 @@ export const isSidebarOpen = () => appStore.isSidebarOpen;
 export const toggleSidebar = () => appStore.toggleSidebar();
 export const toggleDebugMode = () => appStore.toggleDebugMode();
 export const setDebugMode = (enabled: boolean) => appStore.setDebugMode(enabled);
-export const toggleTopologyOnlyMode = () => appStore.toggleTopologyOnlyMode();
-export const setTopologyOnlyMode = (enabled: boolean) => appStore.setTopologyOnlyMode(enabled);
-export const toggleChatSidebarVisible = () => appStore.toggleChatSidebarVisible();
-export const setChatSidebarVisible = (visible: boolean) => appStore.setChatSidebarVisible(visible);
 export const refreshState = () => appStore.fetchState();
 

dashboard/src/routes/+page.svelte

@@ -18,10 +18,6 @@
 		selectedChatModel,
 	debugMode,
 	toggleDebugMode,
-	topologyOnlyMode,
-	toggleTopologyOnlyMode,
-	chatSidebarVisible,
-	toggleChatSidebarVisible,
 		type DownloadProgress,
 		type PlacementPreview
 	} from '$lib/stores/app.svelte';
@@ -41,8 +37,6 @@
 	const selectedModelId = $derived(selectedPreviewModelId());
 	const loadingPreviews = $derived(isLoadingPreviews());
 const debugEnabled = $derived(debugMode());
-const topologyOnlyEnabled = $derived(topologyOnlyMode());
-const sidebarVisible = $derived(chatSidebarVisible());
 
 	let mounted = $state(false);
 
@@ -478,7 +472,6 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 				
 				const progress = parseDownloadProgress(downloadPayload);
 				if (progress) {
-					// Sum all values across nodes - each node downloads independently
 					totalBytes += progress.totalBytes;
 					downloadedBytes += progress.downloadedBytes;
 					totalSpeed += progress.speed;
@@ -496,17 +489,13 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 			return { isDownloading: false, progress: null, perNode: [] };
 		}
 
-		// ETA = total remaining bytes / total speed across all nodes
-		const remainingBytes = totalBytes - downloadedBytes;
-		const etaMs = totalSpeed > 0 ? (remainingBytes / totalSpeed) * 1000 : 0;
-
 		return {
 			isDownloading: true,
 			progress: {
 				totalBytes,
 				downloadedBytes,
 				speed: totalSpeed,
-				etaMs,
+				etaMs: totalSpeed > 0 ? ((totalBytes - downloadedBytes) / totalSpeed) * 1000 : 0,
 				percentage: totalBytes > 0 ? (downloadedBytes / totalBytes) * 100 : 0,
 				completedFiles,
 				totalFiles,
@@ -587,7 +576,6 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 					
 					const progress = parseDownloadProgress(downloadPayload);
 					if (progress) {
-						// Sum all values across nodes - each node downloads independently
 						totalBytes += progress.totalBytes;
 						downloadedBytes += progress.downloadedBytes;
 						totalSpeed += progress.speed;
@@ -608,17 +596,13 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 			return { isDownloading: false, progress: null, statusText: statusInfo.statusText, perNode: [] };
 		}
 
-		// ETA = total remaining bytes / total speed across all nodes
-		const remainingBytes = totalBytes - downloadedBytes;
-		const etaMs = totalSpeed > 0 ? (remainingBytes / totalSpeed) * 1000 : 0;
-
 		return {
 			isDownloading: true,
 			progress: {
 				totalBytes,
 				downloadedBytes,
 				speed: totalSpeed,
-				etaMs,
+				etaMs: totalSpeed > 0 ? ((totalBytes - downloadedBytes) / totalSpeed) * 1000 : 0,
 				percentage: totalBytes > 0 ? (downloadedBytes / totalBytes) * 100 : 0,
 				completedFiles,
 				totalFiles,
@@ -634,12 +618,10 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 	function getStatusColor(statusText: string): string {
 		switch (statusText) {
 			case 'FAILED': return 'text-red-400';
-			case 'SHUTDOWN': return 'text-gray-400';
 			case 'DOWNLOADING': return 'text-blue-400';
 			case 'LOADING': 
 			case 'WARMING UP': 
-			case 'WAITING':
-			case 'INITIALIZING': return 'text-yellow-400';
+			case 'WAITING': return 'text-yellow-400';
 			case 'RUNNING': return 'text-teal-400';
 			case 'READY': 
 			case 'LOADED': return 'text-green-400';
@@ -662,15 +644,12 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 				if (!r) return null;
 				const [kind] = getTagged(r);
 				const statusMap: Record<string, string> = {
-					RunnerWaitingForInitialization: 'WaitingForInitialization',
-					RunnerInitializingBackend: 'InitializingBackend',
 					RunnerWaitingForModel: 'WaitingForModel',
 					RunnerLoading: 'Loading',
 					RunnerLoaded: 'Loaded',
 					RunnerWarmingUp: 'WarmingUp',
 					RunnerReady: 'Ready',
 					RunnerRunning: 'Running',
-					RunnerShutdown: 'Shutdown',
 					RunnerFailed: 'Failed',
 				};
 				return kind ? statusMap[kind] || null : null;
@@ -681,15 +660,12 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 
 		if (statuses.length === 0) return { statusText: 'UNKNOWN', statusClass: 'inactive' };
 		if (has('Failed')) return { statusText: 'FAILED', statusClass: 'failed' };
-		if (has('Shutdown')) return { statusText: 'SHUTDOWN', statusClass: 'inactive' };
 		if (has('Loading')) return { statusText: 'LOADING', statusClass: 'starting' };
 		if (has('WarmingUp')) return { statusText: 'WARMING UP', statusClass: 'starting' };
 		if (has('Running')) return { statusText: 'RUNNING', statusClass: 'running' };
 		if (has('Ready')) return { statusText: 'READY', statusClass: 'loaded' };
 		if (has('Loaded')) return { statusText: 'LOADED', statusClass: 'loaded' };
 		if (has('WaitingForModel')) return { statusText: 'WAITING', statusClass: 'starting' };
-		if (has('InitializingBackend')) return { statusText: 'INITIALIZING', statusClass: 'starting' };
-		if (has('WaitingForInitialization')) return { statusText: 'INITIALIZING', statusClass: 'starting' };
 
 		return { statusText: 'RUNNING', statusClass: 'active' };
 	}
@@ -1131,47 +1107,16 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 		<div class="shooting-star" style="top: 50%; left: 40%; --duration: 45s; --delay: 30s;"></div>
 	</div>
 
-	{#if !topologyOnlyEnabled}
-	<HeaderNav 
-		showHome={chatStarted} 
-		onHome={handleGoHome} 
-		showSidebarToggle={true}
-		sidebarVisible={sidebarVisible}
-		onToggleSidebar={toggleChatSidebarVisible}
-	/>
-	{/if}
+	<HeaderNav showHome={chatStarted} onHome={handleGoHome} />
 
 	<!-- Main Content -->
 	<main class="flex-1 flex overflow-hidden relative">
-		<!-- Left: Conversation History Sidebar (hidden in topology-only mode or when toggled off) -->
-		{#if !topologyOnlyEnabled && sidebarVisible}
+		<!-- Left: Conversation History Sidebar (always visible) -->
 		<div class="w-80 flex-shrink-0 border-r border-exo-yellow/10">
 			<ChatSidebar class="h-full" />
 		</div>
-		{/if}
 
-		{#if topologyOnlyEnabled}
-			<!-- TOPOLOGY ONLY MODE: Full-screen topology -->
-			<div class="flex-1 flex flex-col min-h-0 min-w-0 p-4" in:fade={{ duration: 300 }}>
-				<div class="flex-1 relative bg-exo-dark-gray/40 rounded-lg overflow-hidden">
-					<TopologyGraph class="w-full h-full" highlightedNodes={highlightedNodes()} />
-					<!-- Exit topology-only mode button -->
-					<button
-						type="button"
-						onclick={toggleTopologyOnlyMode}
-						class="absolute bottom-4 right-4 p-2 rounded border border-exo-yellow/30 bg-exo-dark-gray/80 hover:border-exo-yellow/50 hover:bg-exo-dark-gray transition-colors cursor-pointer backdrop-blur-sm"
-						title="Exit topology only mode"
-					>
-						<svg class="w-5 h-5 text-exo-yellow" fill="none" viewBox="0 0 24 24" stroke="currentColor" stroke-width="2">
-							<circle cx="12" cy="5" r="2" fill="currentColor" />
-							<circle cx="5" cy="19" r="2" fill="currentColor" />
-							<circle cx="19" cy="19" r="2" fill="currentColor" />
-							<path stroke-linecap="round" d="M12 7v5m0 0l-5 5m5-5l5 5" />
-						</svg>
-					</button>
-				</div>
-			</div>
-		{:else if !chatStarted}
+		{#if !chatStarted}
 			<!-- WELCOME STATE: Topology + Instance Controls (no left sidebar for cleaner look) -->
 			<div class="flex-1 flex overflow-visible relative" in:fade={{ duration: 300 }} out:fade={{ duration: 200 }}>
 				
@@ -1355,15 +1300,14 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 																			{:else}
 																				{#each nodeProg.progress.files as f}
 																					{@const filePercent = Math.min(100, Math.max(0, f.percentage ?? 0))}
-																					{@const isFileComplete = filePercent >= 100}
 																					<div class="rounded border border-exo-medium-gray/30 bg-exo-black/40 p-2">
 																						<div class="flex items-center justify-between text-[10px] font-mono text-exo-light-gray/90">
 																							<span class="truncate pr-2">{f.name}</span>
-																							<span class={isFileComplete ? 'text-green-400' : 'text-white/80'}>{filePercent.toFixed(1)}%</span>
+																							<span class="text-white/80">{filePercent.toFixed(1)}%</span>
 																						</div>
 																						<div class="relative h-1 bg-exo-black/60 rounded-sm overflow-hidden mt-1">
 																							<div 
-																								class="absolute inset-y-0 left-0 bg-gradient-to-r {isFileComplete ? 'from-green-500 to-green-400' : 'from-exo-yellow to-exo-yellow/70'} transition-all duration-300"
+																								class="absolute inset-y-0 left-0 bg-gradient-to-r from-exo-yellow to-exo-yellow/70 transition-all duration-300"
 																								style="width: {filePercent.toFixed(1)}%"
 																							></div>
 																						</div>
@@ -1667,13 +1611,13 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 					in:fade={{ duration: 300, delay: 100 }}
 				>
 					<div class="flex-1 overflow-y-auto px-8 py-6" bind:this={chatScrollRef}>
-						<div class="max-w-7xl mx-auto">
+						<div class="max-w-3xl mx-auto">
 							<ChatMessages scrollParent={chatScrollRef} />
 						</div>
 					</div>
 					
 					<div class="flex-shrink-0 px-8 pb-6 pt-4 bg-gradient-to-t from-exo-black via-exo-black to-transparent">
-						<div class="max-w-7xl mx-auto">
+						<div class="max-w-3xl mx-auto">
 							<ChatForm placeholder="Ask anything" showModelSelector={true} />
 						</div>
 					</div>
@@ -1711,7 +1655,7 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 							<!-- Panel Header -->
 							<div class="flex items-center gap-2 mb-4">
 								<div class="w-2 h-2 bg-exo-yellow rounded-full shadow-[0_0_8px_rgba(255,215,0,0.6)] animate-pulse"></div>
-								<h3 class="text-xs text-exo-yellow font-mono tracking-[0.2em] uppercase">Instances</h3>
+								<h3 class="text-sm text-exo-yellow font-mono tracking-[0.2em] uppercase">Instances</h3>
 								<div class="flex-1 h-px bg-gradient-to-r from-exo-yellow/30 to-transparent"></div>
 							</div>
 								<div class="space-y-3 max-h-72 overflow-y-auto pr-1">
@@ -1757,28 +1701,28 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 											<div class="flex justify-between items-start mb-2 pl-2">
 												<div class="flex items-center gap-2">
 													<div class="w-1.5 h-1.5 {isDownloading ? 'bg-blue-400 animate-pulse' : isFailed ? 'bg-red-400' : isLoading ? 'bg-yellow-400 animate-pulse' : isReady ? 'bg-green-400' : 'bg-teal-400'} rounded-full shadow-[0_0_6px_currentColor]"></div>
-													<span class="text-exo-light-gray font-mono text-sm tracking-wider">{id.slice(0, 8).toUpperCase()}</span>
+													<span class="text-exo-light-gray font-mono text-xs tracking-wider">{id.slice(0, 8).toUpperCase()}</span>
 												</div>
 												<button 
 													onclick={() => deleteInstance(id)}
-													class="text-xs px-2 py-1 font-mono tracking-wider uppercase border border-red-500/30 text-red-400 hover:bg-red-500/20 hover:text-red-400 hover:border-red-500/50 transition-all duration-200 cursor-pointer"
+													class="text-xs px-2 py-1 font-mono tracking-wider uppercase border border-red-500/30 text-red-400/80 hover:bg-red-500/20 hover:text-red-400 hover:border-red-500/50 transition-all duration-200 cursor-pointer"
 												>
 													DELETE
 												</button>
 												</div>
 												<div class="pl-2">
-													<div class="text-exo-yellow text-xs font-mono tracking-wide truncate">{getInstanceModelId(instance)}</div>
+													<div class="text-exo-yellow text-sm font-mono tracking-wide truncate">{getInstanceModelId(instance)}</div>
 													<div class="text-white/60 text-xs font-mono">Strategy: <span class="text-white/80">{instanceInfo.sharding} ({instanceInfo.instanceType})</span></div>
 														{#if instanceModelId && instanceModelId !== 'Unknown' && instanceModelId !== 'Unknown Model'}
 															<a
-																class="inline-flex items-center gap-1 text-[11px] text-white/60 hover:text-exo-yellow transition-colors mt-1"
+																class="inline-flex items-center gap-1 text-[10px] text-white/60 hover:text-exo-yellow transition-colors mt-0.5"
 																href={`https://huggingface.co/${instanceModelId}`}
 																target="_blank"
 																rel="noreferrer noopener"
 																aria-label="View model on Hugging Face"
 															>
 																<span>Hugging Face</span>
-																<svg class="w-3.5 h-3.5" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+																<svg class="w-3 h-3" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
 																	<path d="M14 3h7v7"/>
 																	<path d="M10 14l11-11"/>
 																	<path d="M21 14v6a1 1 0 0 1-1 1h-16a1 1 0 0 1-1-1v-16a1 1 0 0 1 1-1h6"/>
@@ -1789,84 +1733,68 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 														<div class="text-white/60 text-xs font-mono">{instanceInfo.nodeNames.join(', ')}</div>
 													{/if}
 													{#if debugEnabled && instanceConnections.length > 0}
-													<div class="mt-2 space-y-1">
-														{#each instanceConnections as conn}
-															<div class="text-[11px] leading-snug font-mono text-white/70">
-																<span>{conn.from} -> {conn.to}: {conn.ip}</span>
-																<span class="{conn.missingIface ? 'text-red-400' : 'text-white/60'}"> ({conn.ifaceLabel})</span>
+														<div class="mt-1 space-y-0.5">
+															{#each instanceConnections as conn}
+																<div class="text-[10px] leading-snug font-mono text-white/70">
+																	<span>{conn.from} -> {conn.to}: {conn.ip}</span>
+																	<span class="{conn.missingIface ? 'text-red-400' : 'text-white/60'}"> ({conn.ifaceLabel})</span>
+																</div>
+															{/each}
+														</div>
+													{/if}
+													
+													<!-- Download Progress -->
+													{#if downloadInfo.isDownloading && downloadInfo.progress}
+														<div class="mt-2 space-y-1">
+															<div class="flex justify-between text-sm font-mono">
+																<span class="text-blue-400">{downloadInfo.progress.percentage.toFixed(1)}%</span>
+																<span class="text-exo-light-gray">{formatBytes(downloadInfo.progress.downloadedBytes)}/{formatBytes(downloadInfo.progress.totalBytes)}</span>
+															</div>
+															<div class="relative h-1 bg-exo-black/60 rounded-sm overflow-hidden">
+																<div 
+																	class="absolute inset-y-0 left-0 bg-gradient-to-r from-blue-500 to-blue-400 transition-all duration-300"
+																	style="width: {downloadInfo.progress.percentage}%"
+																></div>
+															</div>
+															<div class="flex justify-between text-xs font-mono text-exo-light-gray">
+																<span>{formatSpeed(downloadInfo.progress.speed)}</span>
+																<span>ETA: {formatEta(downloadInfo.progress.etaMs)}</span>
+																<span>{downloadInfo.progress.completedFiles}/{downloadInfo.progress.totalFiles} files</span>
 															</div>
-														{/each}
-													</div>
-												{/if}
-												
-												<!-- Download Progress -->
-												{#if downloadInfo.isDownloading && downloadInfo.progress}
-													<div class="mt-2 space-y-1">
-														<div class="flex justify-between text-xs font-mono">
-															<span class="text-blue-400">{downloadInfo.progress.percentage.toFixed(1)}%</span>
-															<span class="text-exo-light-gray">{formatBytes(downloadInfo.progress.downloadedBytes)}/{formatBytes(downloadInfo.progress.totalBytes)}</span>
 														</div>
-														<div class="relative h-1.5 bg-exo-black/60 rounded-sm overflow-hidden">
-															<div 
-																class="absolute inset-y-0 left-0 bg-gradient-to-r from-blue-500 to-blue-400 transition-all duration-300"
-																style="width: {downloadInfo.progress.percentage}%"
-															></div>
-														</div>
-														<div class="flex justify-between text-xs font-mono text-exo-light-gray">
-															<span>{formatSpeed(downloadInfo.progress.speed)}</span>
-															<span>ETA: {formatEta(downloadInfo.progress.etaMs)}</span>
-															<span>{downloadInfo.progress.completedFiles}/{downloadInfo.progress.totalFiles} files</span>
-														</div>
-													</div>
-													{#if downloadInfo.perNode.length > 0}
-														<div class="mt-2 space-y-2 max-h-48 overflow-y-auto pr-1">
-															{#each downloadInfo.perNode as nodeProg}
-																{@const nodePercent = Math.min(100, Math.max(0, nodeProg.progress.percentage))}
-																{@const isExpanded = instanceDownloadExpandedNodes.has(nodeProg.nodeId)}
-																<div class="rounded border border-exo-medium-gray/40 bg-exo-black/30 p-2">
-																	<button
-																		type="button"
-																		class="w-full text-left space-y-1.5"
-																		onclick={() => toggleInstanceDownloadDetails(nodeProg.nodeId)}
-																	>
-																		<div class="flex items-center justify-between text-[11px] font-mono text-exo-light-gray">
+														{#if downloadInfo.perNode.length > 0}
+															<div class="mt-2 space-y-1.5 max-h-48 overflow-y-auto pr-1">
+																{#each downloadInfo.perNode as nodeProg}
+																	<div class="rounded border border-exo-medium-gray/40 bg-exo-black/30 p-2">
+																		<div class="flex items-center justify-between text-[11px] font-mono text-exo-light-gray mb-1">
 																			<span class="text-white/80 truncate pr-2">{nodeProg.nodeName}</span>
-																			<span class="flex items-center gap-1 text-blue-300">
-																				{nodePercent.toFixed(1)}%
-																				<svg class="w-3 h-3 text-exo-light-gray" viewBox="0 0 20 20" fill="none" stroke="currentColor" stroke-width="2">
-																					<path d="M6 8l4 4 4-4" class={isExpanded ? 'transform rotate-180 origin-center transition-transform duration-150' : 'transition-transform duration-150'}></path>
-																				</svg>
-																			</span>
+																			<span class="text-blue-300">{Math.min(100, Math.max(0, nodeProg.progress.percentage)).toFixed(1)}%</span>
 																		</div>
-																		<div class="relative h-1.5 bg-exo-black/60 rounded-sm overflow-hidden">
+																		<div class="relative h-1 bg-exo-black/60 rounded-sm overflow-hidden mb-1.5">
 																			<div 
-																				class="absolute inset-y-0 left-0 bg-gradient-to-r from-blue-500 to-blue-400 transition-all duration-300"
-																				style="width: {nodePercent.toFixed(1)}%"
+																				class="absolute inset-y-0 left-0 bg-blue-500/80 transition-all duration-300"
+																				style="width: {Math.min(100, Math.max(0, nodeProg.progress.percentage)).toFixed(1)}%"
 																			></div>
 																		</div>
-																		<div class="flex items-center justify-between text-[11px] font-mono text-exo-light-gray">
+																		<div class="flex items-center justify-between text-[11px] font-mono text-exo-light-gray mb-1">
 																			<span>{formatBytes(nodeProg.progress.downloadedBytes)} / {formatBytes(nodeProg.progress.totalBytes)}</span>
 																			<span>{formatSpeed(nodeProg.progress.speed)} • ETA {formatEta(nodeProg.progress.etaMs)}</span>
 																		</div>
-																	</button>
-
-																	{#if isExpanded}
-																		<div class="mt-2 space-y-1.5">
-																			{#if nodeProg.progress.files.length === 0}
-																				<div class="text-[11px] font-mono text-exo-light-gray/70">No file details reported.</div>
-																			{:else}
-																				{#each nodeProg.progress.files as f}
-																					{@const filePercent = Math.min(100, Math.max(0, f.percentage ?? 0))}
-																					{@const isFileComplete = filePercent >= 100}
-																					<div class="rounded border border-exo-medium-gray/30 bg-exo-black/40 p-2">
-																						<div class="flex items-center justify-between text-[10px] font-mono text-exo-light-gray/90">
+																	{#if nodeProg.progress.files.length > 0}
+																		{@const inProgressFiles = nodeProg.progress.files.filter(f => (f.percentage ?? 0) < 100)}
+																		{@const completedFiles = nodeProg.progress.files.filter(f => (f.percentage ?? 0) >= 100)}
+																		{#if inProgressFiles.length > 0}
+																			<div class="space-y-1">
+																				{#each inProgressFiles as f}
+																					<div class="text-[10px] font-mono text-exo-light-gray/80">
+																						<div class="flex items-center justify-between">
 																							<span class="truncate pr-2">{f.name}</span>
-																							<span class={isFileComplete ? 'text-green-400' : 'text-white/80'}>{filePercent.toFixed(1)}%</span>
+																							<span class="text-white/70">{Math.min(100, Math.max(0, f.percentage)).toFixed(1)}%</span>
 																						</div>
-																						<div class="relative h-1 bg-exo-black/60 rounded-sm overflow-hidden mt-1">
+																						<div class="relative h-1 bg-exo-black/50 rounded-sm overflow-hidden mt-0.5">
 																							<div 
-																								class="absolute inset-y-0 left-0 bg-gradient-to-r {isFileComplete ? 'from-green-500 to-green-400' : 'from-exo-yellow to-exo-yellow/70'} transition-all duration-300"
-																								style="width: {filePercent.toFixed(1)}%"
+																								class="absolute inset-y-0 left-0 bg-gradient-to-r from-exo-yellow to-exo-yellow/70"
+																								style="width: {Math.min(100, Math.max(0, f.percentage)).toFixed(1)}%"
 																							></div>
 																						</div>
 																						<div class="flex items-center justify-between text-[10px] text-exo-light-gray/70 mt-0.5">
@@ -1875,17 +1803,27 @@ function toggleInstanceDownloadDetails(nodeId: string): void {
 																						</div>
 																					</div>
 																				{/each}
-																			{/if}
-																		</div>
+																			</div>
+																		{/if}
+																		{#if completedFiles.length > 0}
+																			<div class="pt-1 space-y-0.5">
+																				{#each completedFiles as f}
+																					<div class="text-[10px] font-mono text-exo-light-gray/70 flex items-center justify-between">
+																						<span class="truncate pr-2">{f.name}</span>
+																						<span class="text-white/60">100%</span>
+																					</div>
+																				{/each}
+																			</div>
+																		{/if}
 																	{/if}
-																</div>
-															{/each}
-														</div>
+																	</div>
+																{/each}
+															</div>
+														{/if}
+														<div class="text-sm text-blue-400 font-mono tracking-wider mt-1">DOWNLOADING</div>
+													{:else}
+														<div class="text-sm {getStatusColor(downloadInfo.statusText)} font-mono tracking-wider mt-1">{downloadInfo.statusText}</div>
 													{/if}
-													<div class="text-xs text-blue-400 font-mono tracking-wider mt-1">DOWNLOADING</div>
-												{:else}
-													<div class="text-xs {getStatusColor(downloadInfo.statusText)} font-mono tracking-wider mt-1">{downloadInfo.statusText}</div>
-												{/if}
 												</div>
 											</div>
 										</div>

dashboard/src/routes/downloads/+page.svelte

@@ -345,19 +345,13 @@
 							<div class="rounded border border-exo-medium-gray/30 bg-exo-dark-gray/60 p-3 space-y-2">
 								<div class="flex items-center justify-between gap-3">
 									<div class="min-w-0 space-y-0.5">
-										<div 
-											class="text-xs font-mono text-white truncate"
-											title={model.prettyName ?? model.modelId}
-										>{model.prettyName ?? model.modelId}</div>
-										<div 
-											class="text-[10px] text-exo-light-gray font-mono truncate"
-											title={model.modelId}
-										>{model.modelId}</div>
-										{#if model.status !== 'completed'}
-											<div class="text-[11px] text-exo-light-gray font-mono">
-												{formatBytes(model.downloadedBytes)} / {formatBytes(model.totalBytes)}
-											</div>
-										{/if}
+										<div class="text-sm font-mono text-white truncate">{model.prettyName ?? model.modelId}</div>
+										<div class="text-[11px] text-exo-light-gray font-mono truncate">
+											{model.modelId}
+										</div>
+										<div class="text-[11px] text-exo-light-gray font-mono">
+											{formatBytes(model.downloadedBytes)} / {formatBytes(model.totalBytes)}
+										</div>
 									</div>
 									<div class="flex items-center gap-2">
 										<span class="text-xs font-mono {pct >= 100 ? 'text-green-400' : pct <= 0 ? 'text-red-400' : 'text-exo-yellow'}">
@@ -432,14 +426,14 @@
 <style>
 	.downloads-grid {
 		display: grid;
-		grid-template-columns: repeat(auto-fill, minmax(320px, 1fr));
+		grid-template-columns: repeat(auto-fill, minmax(260px, 1fr));
 	}
 	@media (min-width: 1024px) {
 		.downloads-grid {
 			grid-template-columns: repeat(3, minmax(0, 1fr));
 		}
 	}
-	@media (min-width: 1600px) {
+	@media (min-width: 1440px) {
 		.downloads-grid {
 			grid-template-columns: repeat(4, minmax(0, 1fr));
 		}

pyproject.toml

@@ -29,12 +29,10 @@ dependencies = [
     "exo_pyo3_bindings", # rust bindings
     "anyio==4.11.0",
     "bidict>=0.23.1",
-    "mlx>=0.30.1; sys_platform == 'darwin'",
-    "mlx[cpu]>=0.30.1; sys_platform == 'linux'",
+    "mlx>=0.30.1",
     "mlx-lm>=0.28.3",
     "tiktoken>=0.12.0", # required for kimi k2 tokenizer
     "hypercorn>=0.18.0",
-    "openai-harmony>=0.0.8",
 ]
 
 [project.scripts]

src/exo/main.py

@@ -1,6 +1,5 @@
 import argparse
 import multiprocessing as mp
-import os
 import signal
 from dataclasses import dataclass, field
 from typing import Self
@@ -195,7 +194,6 @@ def main():
     # TODO: Refactor the current verbosity system
     logger_setup(EXO_LOG, args.verbosity)
     logger.info("Starting EXO")
-    logger.info(f"EXO_LIBP2P_NAMESPACE: {os.getenv('EXO_LIBP2P_NAMESPACE')}")
 
     node = anyio.run(Node.create, args)
     anyio.run(node.run)

src/exo/master/api.py

@@ -13,12 +13,6 @@ from hypercorn.asyncio import serve  # pyright: ignore[reportUnknownVariableType
 from hypercorn.config import Config
 from hypercorn.typing import ASGIFramework
 from loguru import logger
-from openai_harmony import (  # pyright: ignore[reportMissingTypeStubs]
-    HarmonyEncodingName,
-    Role,
-    StreamableParser,
-    load_harmony_encoding,
-)
 
 from exo.master.placement import place_instance as get_instance_placements
 from exo.shared.apply import apply
@@ -27,13 +21,11 @@ from exo.shared.logging import InterceptLogger
 from exo.shared.models.model_cards import MODEL_CARDS
 from exo.shared.models.model_meta import get_model_meta
 from exo.shared.types.api import (
-    ChatCompletionChoice,
     ChatCompletionMessage,
     ChatCompletionResponse,
     CreateInstanceParams,
     CreateInstanceResponse,
     DeleteInstanceResponse,
-    FinishReason,
     ModelList,
     ModelListModel,
     PlaceInstanceParams,
@@ -64,7 +56,7 @@ from exo.utils.channels import Receiver, Sender, channel
 from exo.utils.dashboard_path import find_dashboard
 from exo.utils.event_buffer import OrderedBuffer
 
-encoding = load_harmony_encoding(HarmonyEncodingName.HARMONY_GPT_OSS)
+HIDE_THINKING = False
 
 
 def chunk_to_response(
@@ -169,9 +161,7 @@ class API:
         self.app.delete("/instance/{instance_id}")(self.delete_instance)
         self.app.get("/models")(self.get_models)
         self.app.get("/v1/models")(self.get_models)
-        self.app.post("/v1/chat/completions", response_model=None)(
-            self.chat_completions
-        )
+        self.app.post("/v1/chat/completions")(self.chat_completions)
         self.app.get("/state")(lambda: self.state)
         self.app.get("/events")(lambda: self._event_log)
 
@@ -187,32 +177,17 @@ class API:
         return CreateInstanceResponse(
             message="Command received.",
             command_id=command.command_id,
-            model_meta=command.model_meta,
         )
 
     async def create_instance(
         self, payload: CreateInstanceParams
     ) -> CreateInstanceResponse:
-        instance = payload.instance
-        model_meta = await resolve_model_meta(instance.shard_assignments.model_id)
-        required_memory = model_meta.storage_size
-        available_memory = self._calculate_total_available_memory()
-
-        if required_memory > available_memory:
-            raise HTTPException(
-                status_code=400,
-                detail=f"Insufficient memory to create instance. Required: {required_memory.in_gb:.1f}GB, Available: {available_memory.in_gb:.1f}GB",
-            )
-
-        command = CreateInstance(
-            instance=instance,
-        )
+        command = CreateInstance(instance=payload.instance)
         await self._send(command)
 
         return CreateInstanceResponse(
             message="Command received.",
             command_id=command.command_id,
-            model_meta=model_meta,
         )
 
     async def get_placement(
@@ -377,52 +352,32 @@ class API:
             instance_id=instance_id,
         )
 
-    async def _process_gpt_oss(self, token_chunks: Receiver[TokenChunk]):
-        stream = StreamableParser(encoding, role=Role.ASSISTANT)
-        thinking = False
-
-        async for chunk in token_chunks:
-            stream.process(chunk.token_id)
-
-            delta = stream.last_content_delta
-            ch = stream.current_channel
-
-            if ch == "analysis" and not thinking:
-                thinking = True
-                yield chunk.model_copy(update={"text": "<think>"})
-
-            if ch != "analysis" and thinking:
-                thinking = False
-                yield chunk.model_copy(update={"text": "</think>"})
-
-            if delta:
-                yield chunk.model_copy(update={"text": delta})
-
-            if chunk.finish_reason is not None:
-                if thinking:
-                    yield chunk.model_copy(update={"text": "</think>"})
-                yield chunk
-                break
-
-    async def _chat_chunk_stream(
-        self, command_id: CommandId, parse_gpt_oss: bool
-    ) -> AsyncGenerator[TokenChunk, None]:
-        """Yield `TokenChunk`s for a given command until completion."""
+    async def _generate_chat_stream(
+        self, command_id: CommandId
+    ) -> AsyncGenerator[str, None]:
+        """Generate chat completion stream as JSON strings."""
 
         try:
             self._chat_completion_queues[command_id], recv = channel[TokenChunk]()
 
+            is_thinking = False
             with recv as token_chunks:
-                if parse_gpt_oss:
-                    async for chunk in self._process_gpt_oss(token_chunks):
-                        yield chunk
-                        if chunk.finish_reason is not None:
-                            break
-                else:
-                    async for chunk in token_chunks:
-                        yield chunk
-                        if chunk.finish_reason is not None:
-                            break
+                async for chunk in token_chunks:
+                    if HIDE_THINKING:
+                        if chunk.text == "<think>":
+                            is_thinking = True
+                        if chunk.text == "</think>":
+                            is_thinking = False
+                    chunk_response: ChatCompletionResponse = chunk_to_response(
+                        chunk, command_id
+                    )
+                    if not (is_thinking and HIDE_THINKING):
+                        logger.debug(f"chunk_response: {chunk_response}")
+                        yield f"data: {chunk_response.model_dump_json()}\n\n"
+
+                    if chunk.finish_reason is not None:
+                        yield "data: [DONE]\n\n"
+                        break
 
         except anyio.get_cancelled_exc_class():
             # TODO: TaskCancelled
@@ -437,59 +392,6 @@ class API:
             await self._send(command)
             del self._chat_completion_queues[command_id]
 
-    async def _generate_chat_stream(
-        self, command_id: CommandId, parse_gpt_oss: bool
-    ) -> AsyncGenerator[str, None]:
-        """Generate chat completion stream as JSON strings."""
-
-        async for chunk in self._chat_chunk_stream(command_id, parse_gpt_oss):
-            chunk_response: ChatCompletionResponse = chunk_to_response(
-                chunk, command_id
-            )
-            logger.debug(f"chunk_response: {chunk_response}")
-
-            yield f"data: {chunk_response.model_dump_json()}\n\n"
-
-            if chunk.finish_reason is not None:
-                yield "data: [DONE]\n\n"
-
-    async def _collect_chat_completion(
-        self, command_id: CommandId, parse_gpt_oss: bool
-    ) -> ChatCompletionResponse:
-        """Collect all token chunks for a chat completion and return a single response."""
-
-        text_parts: list[str] = []
-        model: str | None = None
-        finish_reason: FinishReason | None = None
-
-        async for chunk in self._chat_chunk_stream(command_id, parse_gpt_oss):
-            if model is None:
-                model = chunk.model
-
-            text_parts.append(chunk.text)
-
-            if chunk.finish_reason is not None:
-                finish_reason = chunk.finish_reason
-
-        combined_text = "".join(text_parts)
-        assert model is not None
-
-        return ChatCompletionResponse(
-            id=command_id,
-            created=int(time.time()),
-            model=model,
-            choices=[
-                ChatCompletionChoice(
-                    index=0,
-                    message=ChatCompletionMessage(
-                        role="assistant",
-                        content=combined_text,
-                    ),
-                    finish_reason=finish_reason,
-                )
-            ],
-        )
-
     async def _trigger_notify_user_to_download_model(self, model_id: str) -> None:
         logger.warning(
             "TODO: we should send a notification to the user to download the model"
@@ -497,12 +399,10 @@ class API:
 
     async def chat_completions(
         self, payload: ChatCompletionTaskParams
-    ) -> ChatCompletionResponse | StreamingResponse:
-        """Handle chat completions, supporting both streaming and non-streaming responses."""
+    ) -> StreamingResponse:
+        """Handle chat completions with proper streaming response."""
         model_meta = await resolve_model_meta(payload.model)
         payload.model = model_meta.model_id
-        parse_gpt_oss = "gpt-oss" in model_meta.model_id.lower()
-        logger.info(f"{parse_gpt_oss=}")
 
         if not any(
             instance.shard_assignments.model_id == payload.model
@@ -517,13 +417,10 @@ class API:
             request_params=payload,
         )
         await self._send(command)
-        if payload.stream:
-            return StreamingResponse(
-                self._generate_chat_stream(command.command_id, parse_gpt_oss),
-                media_type="text/event-stream",
-            )
-
-        return await self._collect_chat_completion(command.command_id, parse_gpt_oss)
+        return StreamingResponse(
+            self._generate_chat_stream(command.command_id),
+            media_type="text/event-stream",
+        )
 
     def _calculate_total_available_memory(self) -> Memory:
         """Calculate total available memory across all nodes in bytes."""
@@ -545,8 +442,6 @@ class API:
                     name=card.name,
                     description=card.description,
                     tags=card.tags,
-                    storage_size_megabytes=int(card.metadata.storage_size.in_mb),
-                    supports_tensor=card.metadata.supports_tensor,
                 )
                 for card in MODEL_CARDS.values()
             ]
@@ -563,7 +458,7 @@ class API:
         async with create_task_group() as tg:
             self._tg = tg
             logger.info("Starting API")
-            tg.start_soon(self._apply_state)
+            tg.start_soon(self._applystate)
             tg.start_soon(self._pause_on_new_election)
             print_startup_banner(self.port)
             await serve(
@@ -575,7 +470,7 @@ class API:
         self.command_sender.close()
         self.global_event_receiver.close()
 
-    async def _apply_state(self):
+    async def _applystate(self):
         with self.global_event_receiver as events:
             async for f_event in events:
                 if f_event.origin != self.session_id.master_node_id:

src/exo/master/placement.py

@@ -21,7 +21,6 @@ from exo.shared.types.commands import (
 )
 from exo.shared.types.events import Event, InstanceCreated, InstanceDeleted
 from exo.shared.types.memory import Memory
-from exo.shared.types.models import ModelId
 from exo.shared.types.topology import NodeInfo
 from exo.shared.types.worker.instances import (
     Instance,
@@ -30,7 +29,6 @@ from exo.shared.types.worker.instances import (
     MlxJacclInstance,
     MlxRingInstance,
 )
-from exo.shared.types.worker.shards import Sharding
 
 
 def random_ephemeral_port() -> int:
@@ -67,28 +65,6 @@ def place_instance(
     if not cycles_with_sufficient_memory:
         raise ValueError("No cycles found with sufficient memory")
 
-    if command.sharding == Sharding.Tensor:
-        if not command.model_meta.supports_tensor:
-            raise ValueError(
-                f"Requested Tensor sharding but this model does not support tensor parallelism: {command.model_meta.model_id}"
-            )
-        # TODO: the condition here for tensor parallel is not correct, but it works good enough for now.
-        cycles_with_sufficient_memory = [
-            cycle
-            for cycle in cycles_with_sufficient_memory
-            if command.model_meta.hidden_size % len(cycle) == 0
-        ]
-        if not cycles_with_sufficient_memory:
-            raise ValueError(
-                f"No tensor sharding found for model with hidden_size {command.model_meta.hidden_size} candidate cycles"
-            )
-    if command.sharding == Sharding.Pipeline and command.model_meta.model_id == ModelId(
-        "mlx-community/DeepSeek-V3.1-8bit"
-    ):
-        raise ValueError(
-            "Pipeline parallelism is not supported for DeepSeek V3.1 (8-bit)"
-        )
-
     smallest_cycles = get_smallest_cycles(cycles_with_sufficient_memory)
 
     smallest_tb_cycles = [

src/exo/master/placement_utils.py

@@ -217,9 +217,7 @@ def get_mlx_ibv_devices_matrix(
             # Find the IP J uses to talk to I
             for connection_ip, _ in _find_connection_ip(node_j, node_i, cycle_digraph):
                 # This is a local IP on I, which is attached to an interface: find that interface
-                if interface_name := _find_rdma_interface_name_for_ip(
-                    connection_ip, node_i
-                ):
+                if interface_name := _find_interface_name_for_ip(connection_ip, node_i):
                     matrix[i][j] = interface_name
                     logger.info(
                         f"Interface name for {connection_ip} on {node_i.node_id}: {interface_name}"
@@ -250,7 +248,7 @@ def _find_connection_ip(
             yield connection.send_back_multiaddr.ip_address, connection.is_thunderbolt()
 
 
-def _find_rdma_interface_name_for_ip(
+def _find_interface_name_for_ip(
     ip_address: str,
     node_info: NodeInfo,
 ) -> str | None:
@@ -271,7 +269,7 @@ def _find_rdma_interface_name_for_ip(
     return None
 
 
-def _find_interface_name_for_ip(
+def _find_general_interface_name_for_ip(
     ip_address: str,
     node_info: NodeInfo,
 ) -> str | None:
@@ -289,7 +287,6 @@ def _find_interface_name_for_ip(
 def _find_ip_prioritised(
     node: NodeInfo, other_node: NodeInfo, cycle_digraph: Topology
 ) -> str | None:
-    # TODO: Actually prioritize in the correct Ethernet > Wifi > Non-TB > TB order.
     """Find an IP address between nodes with prioritization.
 
     Priority order:
@@ -299,21 +296,35 @@ def _find_ip_prioritised(
     4. Any other IP address
     """
     ips = list(_find_connection_ip(node, other_node, cycle_digraph))
-    # We expect a unique iface -> ip mapping
-    iface_map = {_find_interface_name_for_ip(ip, other_node): ip for ip, _ in ips}
+    interface_names = [
+        _find_general_interface_name_for_ip(ip, other_node) for ip, _ in ips
+    ]
 
-    en0_ip = iface_map.get("en0")
+    en0_ip = next(
+        (
+            ip
+            for (ip, _), interface_name in zip(ips, interface_names)
+            if interface_name == "en0"
+        ),
+        None,
+    )
     if en0_ip:
         return en0_ip
 
-    en1_ip = iface_map.get("en1")
+    en1_ip = next(
+        (
+            ip
+            for (ip, _), interface_name in zip(ips, interface_names)
+            if interface_name == "en1"
+        ),
+        None,
+    )
     if en1_ip:
         return en1_ip
 
     non_thunderbolt_ip = next(
         (ip for (ip, is_thunderbolt) in ips if not is_thunderbolt), None
     )
-
     if non_thunderbolt_ip:
         return non_thunderbolt_ip
 
@@ -339,12 +350,16 @@ def get_mlx_ring_hosts_by_node(
     if world_size == 0:
         return {}
 
+    logger.info(f"[RING3DBG] get_mlx_ring_hosts_by_node: world_size={world_size}, ephemeral_port={ephemeral_port}")
+    logger.info(f"[RING3DBG] cycle node_ids: {[n.node_id for n in selected_cycle]}")
+
     hosts_by_node: dict[NodeId, list[Host]] = {}
 
     for rank, node in enumerate(selected_cycle):
         node_id = node.node_id
         left_rank = (rank - 1) % world_size
         right_rank = (rank + 1) % world_size
+        logger.info(f"[RING3DBG] rank={rank} node_id={node_id} left_rank={left_rank} right_rank={right_rank}")
 
         hosts_for_node: list[Host] = []
 
@@ -359,6 +374,7 @@ def get_mlx_ring_hosts_by_node(
                 continue
 
             connection_ip = _find_ip_prioritised(node, other_node, cycle_digraph)
+            logger.info(f"[RING3DBG] rank={rank} idx={idx} connection_ip={connection_ip}")
             if connection_ip is None:
                 logger.warning(
                     f"Failed to find prioritised connection IP between {node_id} and {other_node.node_id}"
@@ -369,6 +385,7 @@ def get_mlx_ring_hosts_by_node(
 
             hosts_for_node.append(Host(ip=connection_ip, port=ephemeral_port))
 
+        logger.info(f"[RING3DBG] rank={rank} final hosts_for_node={hosts_for_node}")
         hosts_by_node[node_id] = hosts_for_node
 
     return hosts_by_node
@@ -385,14 +402,13 @@ def get_mlx_jaccl_coordinators(
     address in format "X.X.X.X:PORT" per node.
     """
     rank_0_node = selected_cycle[0]
-    logger.debug(f"Selecting coordinator from rank 0 node: {rank_0_node.node_id}")
+    logger.info(f"Selecting coordinator from rank 0 node: {rank_0_node.node_id}")
 
     def get_ip_for_node(n: NodeInfo) -> str:
         if n.node_id == rank_0_node.node_id:
             return "0.0.0.0"
 
-        ip = _find_ip_prioritised(n, rank_0_node, cycle_digraph)
-        if ip:
+        for ip, _ in _find_connection_ip(n, rank_0_node, cycle_digraph):
             return ip
 
         logger.warning(

src/exo/master/tests/test_master.py

@@ -123,8 +123,6 @@ async def test_master():
                             pretty_name="Llama 3.2 1B",
                             n_layers=16,
                             storage_size=Memory.from_bytes(678948),
-                            hidden_size=7168,
-                            supports_tensor=True,
                         ),
                         sharding=Sharding.Pipeline,
                         instance_meta=InstanceMeta.MlxRing,
@@ -181,8 +179,6 @@ async def test_master():
                         pretty_name="Llama 3.2 1B",
                         n_layers=16,
                         storage_size=Memory.from_bytes(678948),
-                        hidden_size=7168,
-                        supports_tensor=True,
                     ),
                     device_rank=0,
                     world_size=1,

src/exo/master/tests/test_placement.py

@@ -50,8 +50,6 @@ def model_meta() -> ModelMetadata:
         storage_size=Memory.from_kb(1000),
         pretty_name="Test Model",
         n_layers=10,
-        hidden_size=30,
-        supports_tensor=True,
     )
 
 
@@ -142,8 +140,6 @@ def test_get_instance_placements_one_node_exact_fit(
             storage_size=Memory.from_kb(1000),
             pretty_name="Test Model",
             n_layers=10,
-            hidden_size=1000,
-            supports_tensor=True,
         ),
     )
     placements = place_instance(cic, topology, {})
@@ -169,8 +165,6 @@ def test_get_instance_placements_one_node_fits_with_extra_memory(
             storage_size=Memory.from_kb(1000),
             pretty_name="Test Model",
             n_layers=10,
-            hidden_size=1000,
-            supports_tensor=True,
         ),
     )
     placements = place_instance(cic, topology, {})
@@ -196,8 +190,6 @@ def test_get_instance_placements_one_node_not_fit(
             storage_size=Memory.from_kb(1001),
             pretty_name="Test Model",
             n_layers=10,
-            hidden_size=1000,
-            supports_tensor=True,
         ),
     )
 

src/exo/master/tests/test_placement_utils.py

@@ -198,8 +198,6 @@ def test_get_shard_assignments(
         pretty_name="Test Model",
         n_layers=total_layers,
         storage_size=Memory.from_kb(1000),
-        hidden_size=1000,
-        supports_tensor=True,
     )
     cycles = topology.get_cycles()
     selected_cycle = cycles[0]

src/exo/shared/models/model_cards.py

@@ -51,8 +51,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="DeepSeek V3.1 (4-bit)",
             storage_size=Memory.from_gb(378),
             n_layers=61,
-            hidden_size=7168,
-            supports_tensor=True,
         ),
     ),
     "deepseek-v3.1-8bit": ModelCard(
@@ -66,8 +64,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="DeepSeek V3.1 (8-bit)",
             storage_size=Memory.from_gb(713),
             n_layers=61,
-            hidden_size=7168,
-            supports_tensor=True,
         ),
     ),
     # "deepseek-v3.2": ModelCard(
@@ -139,8 +135,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Kimi K2 Instruct (4-bit)",
             storage_size=Memory.from_gb(578),
             n_layers=61,
-            hidden_size=7168,
-            supports_tensor=True,
         ),
     ),
     "kimi-k2-thinking": ModelCard(
@@ -154,8 +148,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Kimi K2 Thinking (4-bit)",
             storage_size=Memory.from_gb(658),
             n_layers=61,
-            hidden_size=7168,
-            supports_tensor=True,
         ),
     ),
     # llama-3.1
@@ -170,38 +162,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Llama 3.1 8B (4-bit)",
             storage_size=Memory.from_mb(4423),
             n_layers=32,
-            hidden_size=4096,
-            supports_tensor=True,
-        ),
-    ),
-    "llama-3.1-8b-8bit": ModelCard(
-        short_id="llama-3.1-8b-8bit",
-        model_id=ModelId("mlx-community/Meta-Llama-3.1-8B-Instruct-8bit"),
-        name="Llama 3.1 8B (8-bit)",
-        description="""Llama 3.1 is a large language model trained on the Llama 3.1 dataset.""",
-        tags=[],
-        metadata=ModelMetadata(
-            model_id=ModelId("mlx-community/Meta-Llama-3.1-8B-Instruct-8bit"),
-            pretty_name="Llama 3.1 8B (8-bit)",
-            storage_size=Memory.from_mb(8540),
-            n_layers=32,
-            hidden_size=4096,
-            supports_tensor=True,
-        ),
-    ),
-    "llama-3.1-8b-bf16": ModelCard(
-        short_id="llama-3.1-8b-bf16",
-        model_id=ModelId("mlx-community/Meta-Llama-3.1-8B-Instruct-bf16"),
-        name="Llama 3.1 8B (BF16)",
-        description="""Llama 3.1 is a large language model trained on the Llama 3.1 dataset.""",
-        tags=[],
-        metadata=ModelMetadata(
-            model_id=ModelId("mlx-community/Meta-Llama-3.1-8B-Instruct-bf16"),
-            pretty_name="Llama 3.1 8B (BF16)",
-            storage_size=Memory.from_mb(16100),
-            n_layers=32,
-            hidden_size=4096,
-            supports_tensor=True,
         ),
     ),
     "llama-3.1-70b": ModelCard(
@@ -215,8 +175,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Llama 3.1 70B (4-bit)",
             storage_size=Memory.from_mb(38769),
             n_layers=80,
-            hidden_size=8192,
-            supports_tensor=True,
         ),
     ),
     # llama-3.2
@@ -231,8 +189,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Llama 3.2 1B (4-bit)",
             storage_size=Memory.from_mb(696),
             n_layers=16,
-            hidden_size=2048,
-            supports_tensor=True,
         ),
     ),
     "llama-3.2-3b": ModelCard(
@@ -246,8 +202,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Llama 3.2 3B (4-bit)",
             storage_size=Memory.from_mb(1777),
             n_layers=28,
-            hidden_size=3072,
-            supports_tensor=True,
         ),
     ),
     "llama-3.2-3b-8bit": ModelCard(
@@ -261,8 +215,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Llama 3.2 3B (8-bit)",
             storage_size=Memory.from_mb(3339),
             n_layers=28,
-            hidden_size=3072,
-            supports_tensor=True,
         ),
     ),
     # llama-3.3
@@ -277,8 +229,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Llama 3.3 70B",
             storage_size=Memory.from_mb(38769),
             n_layers=80,
-            hidden_size=8192,
-            supports_tensor=True,
         ),
     ),
     "llama-3.3-70b-8bit": ModelCard(
@@ -292,8 +242,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Llama 3.3 70B (8-bit)",
             storage_size=Memory.from_mb(73242),
             n_layers=80,
-            hidden_size=8192,
-            supports_tensor=True,
         ),
     ),
     "llama-3.3-70b-fp16": ModelCard(
@@ -307,8 +255,20 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Llama 3.3 70B (FP16)",
             storage_size=Memory.from_mb(137695),
             n_layers=80,
-            hidden_size=8192,
-            supports_tensor=True,
+        ),
+    ),
+    # phi-3
+    "phi-3-mini": ModelCard(
+        short_id="phi-3-mini",
+        model_id=ModelId("mlx-community/Phi-3-mini-128k-instruct-4bit"),
+        name="Phi 3 Mini 128k (4-bit)",
+        description="""Phi 3 Mini is a large language model trained on the Phi 3 Mini dataset.""",
+        tags=[],
+        metadata=ModelMetadata(
+            model_id=ModelId("mlx-community/Phi-3-mini-128k-instruct-4bit"),
+            pretty_name="Phi 3 Mini 128k (4-bit)",
+            storage_size=Memory.from_mb(2099),
+            n_layers=32,
         ),
     ),
     # qwen3
@@ -323,8 +283,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Qwen3 0.6B (4-bit)",
             storage_size=Memory.from_mb(327),
             n_layers=28,
-            hidden_size=1024,
-            supports_tensor=False,
         ),
     ),
     "qwen3-0.6b-8bit": ModelCard(
@@ -338,8 +296,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Qwen3 0.6B (8-bit)",
             storage_size=Memory.from_mb(666),
             n_layers=28,
-            hidden_size=1024,
-            supports_tensor=False,
         ),
     ),
     "qwen3-30b": ModelCard(
@@ -353,8 +309,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Qwen3 30B A3B (4-bit)",
             storage_size=Memory.from_mb(16797),
             n_layers=48,
-            hidden_size=2048,
-            supports_tensor=True,
         ),
     ),
     "qwen3-30b-8bit": ModelCard(
@@ -368,68 +322,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Qwen3 30B A3B (8-bit)",
             storage_size=Memory.from_mb(31738),
             n_layers=48,
-            hidden_size=2048,
-            supports_tensor=True,
-        ),
-    ),
-    "qwen3-80b-a3B-4bit": ModelCard(
-        short_id="qwen3-80b-a3B-4bit",
-        model_id=ModelId("mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit"),
-        name="Qwen3 80B A3B (4-bit)",
-        description="""Qwen3 80B""",
-        tags=[],
-        metadata=ModelMetadata(
-            model_id=ModelId("mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit"),
-            pretty_name="Qwen3 80B A3B (4-bit)",
-            storage_size=Memory.from_mb(44800),
-            n_layers=48,
-            hidden_size=2048,
-            supports_tensor=True,
-        ),
-    ),
-    "qwen3-80b-a3B-8bit": ModelCard(
-        short_id="qwen3-80b-a3B-8bit",
-        model_id=ModelId("mlx-community/Qwen3-Next-80B-A3B-Instruct-8bit"),
-        name="Qwen3 80B A3B (8-bit)",
-        description="""Qwen3 80B""",
-        tags=[],
-        metadata=ModelMetadata(
-            model_id=ModelId("mlx-community/Qwen3-Next-80B-A3B-Instruct-8bit"),
-            pretty_name="Qwen3 80B A3B (8-bit)",
-            storage_size=Memory.from_mb(84700),
-            n_layers=48,
-            hidden_size=2048,
-            supports_tensor=True,
-        ),
-    ),
-    "qwen3-80b-a3B-thinking-4bit": ModelCard(
-        short_id="qwen3-80b-a3B-thinking-4bit",
-        model_id=ModelId("mlx-community/Qwen3-Next-80B-A3B-Thinking-4bit"),
-        name="Qwen3 80B A3B Thinking (4-bit)",
-        description="""Qwen3 80B Reasoning model""",
-        tags=[],
-        metadata=ModelMetadata(
-            model_id=ModelId("mlx-community/Qwen3-Next-80B-A3B-Thinking-4bit"),
-            pretty_name="Qwen3 80B A3B (4-bit)",
-            storage_size=Memory.from_mb(84700),
-            n_layers=48,
-            hidden_size=2048,
-            supports_tensor=True,
-        ),
-    ),
-    "qwen3-80b-a3B-thinking-8bit": ModelCard(
-        short_id="qwen3-80b-a3B-thinking-8bit",
-        model_id=ModelId("mlx-community/Qwen3-Next-80B-A3B-Thinking-8bit"),
-        name="Qwen3 80B A3B Thinking (8-bit)",
-        description="""Qwen3 80B Reasoning model""",
-        tags=[],
-        metadata=ModelMetadata(
-            model_id=ModelId("mlx-community/Qwen3-Next-80B-A3B-Thinking-8bit"),
-            pretty_name="Qwen3 80B A3B (8-bit)",
-            storage_size=Memory.from_mb(84700),
-            n_layers=48,
-            hidden_size=2048,
-            supports_tensor=True,
         ),
     ),
     "qwen3-235b-a22b-4bit": ModelCard(
@@ -443,8 +335,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Qwen3 235B A22B (4-bit)",
             storage_size=Memory.from_gb(132),
             n_layers=94,
-            hidden_size=4096,
-            supports_tensor=True,
         ),
     ),
     "qwen3-235b-a22b-8bit": ModelCard(
@@ -458,8 +348,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Qwen3 235B A22B (8-bit)",
             storage_size=Memory.from_gb(250),
             n_layers=94,
-            hidden_size=4096,
-            supports_tensor=True,
         ),
     ),
     "qwen3-coder-480b-a35b-4bit": ModelCard(
@@ -473,8 +361,6 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Qwen3 Coder 480B A35B (4-bit)",
             storage_size=Memory.from_gb(270),
             n_layers=62,
-            hidden_size=6144,
-            supports_tensor=True,
         ),
     ),
     "qwen3-coder-480b-a35b-8bit": ModelCard(
@@ -488,84 +374,77 @@ MODEL_CARDS: dict[str, ModelCard] = {
             pretty_name="Qwen3 Coder 480B A35B (8-bit)",
             storage_size=Memory.from_gb(540),
             n_layers=62,
-            hidden_size=6144,
-            supports_tensor=True,
         ),
     ),
-    # gpt-oss
-    "gpt-oss-120b-MXFP4-Q8": ModelCard(
-        short_id="gpt-oss-120b-MXFP4-Q8",
-        model_id=ModelId("mlx-community/gpt-oss-120b-MXFP4-Q8"),
-        name="GPT-OSS 120B (MXFP4-Q8, MLX)",
-        description="""OpenAI's GPT-OSS 120B is a 117B-parameter Mixture-of-Experts model designed for high-reasoning and general-purpose use; this variant is a 4-bit MLX conversion for Apple Silicon.""",
+    # granite
+    "granite-3.3-2b": ModelCard(
+        short_id="granite-3.3-2b",
+        model_id=ModelId("mlx-community/granite-3.3-2b-instruct-fp16"),
+        name="Granite 3.3 2B (FP16)",
+        description="""Granite-3.3-2B-Instruct is a 2-billion parameter 128K context length language model fine-tuned for improved reasoning and instruction-following capabilities.""",
         tags=[],
         metadata=ModelMetadata(
-            model_id=ModelId("mlx-community/gpt-oss-120b-MXFP4-Q8"),
-            pretty_name="GPT-OSS 120B (MXFP4-Q8, MLX)",
-            storage_size=Memory.from_kb(68_996_301),
-            n_layers=36,
-            hidden_size=2880,
-            supports_tensor=True,
+            model_id=ModelId("mlx-community/granite-3.3-2b-instruct-fp16"),
+            pretty_name="Granite 3.3 2B (FP16)",
+            storage_size=Memory.from_mb(4951),
+            n_layers=40,
         ),
     ),
-    "gpt-oss-20b-4bit": ModelCard(
-        short_id="gpt-oss-20b-4bit",
-        model_id=ModelId("mlx-community/gpt-oss-20b-MXFP4-Q4"),
-        name="GPT-OSS 20B (MXFP4-Q4, MLX)",
-        description="""OpenAI's GPT-OSS 20B is a medium-sized MoE model for lower-latency and local or specialized use cases; this MLX variant uses MXFP4 4-bit quantization.""",
-        tags=[],
-        metadata=ModelMetadata(
-            model_id=ModelId("mlx-community/gpt-oss-20b-MXFP4-Q4"),
-            pretty_name="GPT-OSS 20B (MXFP4-Q4, MLX)",
-            storage_size=Memory.from_kb(11_744_051),
-            n_layers=24,
-            hidden_size=2880,
-            supports_tensor=True,
-        ),
-    ),
-    # Needs to be quantized g32 or g16.
-    "glm-4.5-air-8bit": ModelCard(
-        short_id="glm-4.5-air-8bit",
-        model_id=ModelId("mlx-community/GLM-4.5-Air-8bit"),
-        name="GLM 4.5 Air 8bit",
-        description="""GLM 4.5 Air 8bit""",
-        tags=[],
-        metadata=ModelMetadata(
-            model_id=ModelId("mlx-community/GLM-4.5-Air-8bit"),
-            pretty_name="GLM 4.5 Air 8bit",
-            storage_size=Memory.from_gb(114),
-            n_layers=46,
-            hidden_size=4096,
-            supports_tensor=False,
-        ),
-    ),
-    "glm-4.5-air-bf16": ModelCard(
-        short_id="glm-4.5-air-bf16",
-        model_id=ModelId("mlx-community/GLM-4.5-Air-bf16"),
-        name="GLM 4.5 Air bf16",
-        description="""GLM 4.5 Air bf16""",
-        tags=[],
-        metadata=ModelMetadata(
-            model_id=ModelId("mlx-community/GLM-4.5-Air-bf16"),
-            pretty_name="GLM 4.5 Air bf16",
-            storage_size=Memory.from_gb(214),
-            n_layers=46,
-            hidden_size=4096,
-            supports_tensor=True,
-        ),
-    ),
-    # "devstral-2-123b-instruct-2512-8bit": ModelCard(
-    #     short_id="devstral-2-123b-instruct-2512-8bit",
-    #     model_id=ModelId("mlx-community/Devstral-2-123B-Instruct-2512-8bit"),
-    #     name="Devstral 2 123B Instruct 2512 (8-bit, MLX)",
-    #     description="""Mistral AI's Devstral 2 123B Instruct (2512) is an agentic coding model.""",
+    # "granite-3.3-8b": ModelCard(
+    #     short_id="granite-3.3-8b",
+    #     model_id=ModelId("mlx-community/granite-3.3-8b-instruct-fp16"),
+    #     name="Granite 3.3 8B",
+    #     description="""Granite-3.3-8B-Instruct is a 8-billion parameter 128K context length language model fine-tuned for improved reasoning and instruction-following capabilities.""",
     #     tags=[],
     #     metadata=ModelMetadata(
-    #         model_id=ModelId("mlx-community/Devstral-2-123B-Instruct-2512-8bit"),
-    #         pretty_name="Devstral 2 123B Instruct 2512 (8-bit, MLX)",
-    #         storage_size=Memory.from_kb(133_000_000),
-    #         n_layers=88,
-    #         hidden_size=12288,
+    #         model_id=ModelId("mlx-community/granite-3.3-8b-instruct-fp16"),
+    #         pretty_name="Granite 3.3 8B",
+    #         storage_size=Memory.from_kb(15958720),
+    #         n_layers=40,
+    #     ),
+    # ),
+    # smol-lm
+    # "smol-lm-135m": ModelCard(
+    #     short_id="smol-lm-135m",
+    #     model_id="mlx-community/SmolLM-135M-4bit",
+    #     name="Smol LM 135M",
+    #     description="""SmolLM is a series of state-of-the-art small language models available in three sizes: 135M, 360M, and 1.7B parameters. """,
+    #     tags=[],
+    #     metadata=ModelMetadata(
+    #         model_id=ModelId("mlx-community/SmolLM-135M-4bit"),
+    #         pretty_name="Smol LM 135M",
+    #         storage_size=Memory.from_kb(73940),
+    #         n_layers=30,
+    #     ),
+    # ),
+    # gpt-oss
+    # "gpt-oss-120b-MXFP4-Q8": ModelCard(
+    #     short_id="gpt-oss-120b-MXFP4-Q8",
+    #     model_id=ModelId("mlx-community/gpt-oss-120b-MXFP4-Q8"),
+    #     name="GPT-OSS 120B (MXFP4-Q8, MLX)",
+    #     description="""OpenAI's GPT-OSS 120B is a 117B-parameter Mixture-of-Experts model designed for high-reasoning and general-purpose use; this variant is a 4-bit MLX conversion for Apple Silicon.""",
+    #     tags=[],
+    #     metadata=ModelMetadata(
+    #         model_id=ModelId("mlx-community/gpt-oss-120b-MXFP4-Q8"),
+    #         pretty_name="GPT-OSS 120B (MXFP4-Q8, MLX)",
+    #         storage_size=Memory.from_kb(68_996_301),
+    #         n_layers=36,
+    #         hidden_size=2880,
+    #         supports_tensor=True,
+    #     ),
+    # ),
+    # "gpt-oss-20b-4bit": ModelCard(
+    #     short_id="gpt-oss-20b-4bit",
+    #     model_id=ModelId("mlx-community/gpt-oss-20b-MXFP4-Q4"),
+    #     name="GPT-OSS 20B (MXFP4-Q4, MLX)",
+    #     description="""OpenAI's GPT-OSS 20B is a medium-sized MoE model for lower-latency and local or specialized use cases; this MLX variant uses MXFP4 4-bit quantization.""",
+    #     tags=[],
+    #     metadata=ModelMetadata(
+    #         model_id=ModelId("mlx-community/gpt-oss-20b-MXFP4-Q4"),
+    #         pretty_name="GPT-OSS 20B (MXFP4-Q4, MLX)",
+    #         storage_size=Memory.from_kb(11_744_051),
+    #         n_layers=24,
+    #         hidden_size=2880,
     #         supports_tensor=True,
     #     ),
     # ),

src/exo/shared/models/model_meta.py

@@ -6,7 +6,6 @@ from huggingface_hub import model_info
 from loguru import logger
 from pydantic import BaseModel, Field
 
-from exo.shared.models.model_cards import MODEL_CARDS
 from exo.shared.types.memory import Memory
 from exo.shared.types.models import ModelId, ModelMetadata
 from exo.worker.download.download_utils import (
@@ -26,7 +25,6 @@ class ConfigData(BaseModel):
     n_layers: Annotated[int, Field(ge=0)] | None = None  # Sometimes used
     num_decoder_layers: Annotated[int, Field(ge=0)] | None = None  # Transformer models
     decoder_layers: Annotated[int, Field(ge=0)] | None = None  # Some architectures
-    hidden_size: Annotated[int, Field(ge=0)] | None = None
 
     @property
     def layer_count(self) -> int:
@@ -108,19 +106,10 @@ async def _get_model_meta(model_id: str) -> ModelMetadata:
     config_data = await get_config_data(model_id)
     num_layers = config_data.layer_count
     mem_size_bytes = await get_safetensors_size(model_id)
-    model_card = next(
-        (card for card in MODEL_CARDS.values() if card.model_id == ModelId(model_id)),
-        None,
-    )
 
     return ModelMetadata(
         model_id=ModelId(model_id),
-        pretty_name=model_card.name if model_card is not None else model_id,
+        pretty_name=model_id,
         storage_size=mem_size_bytes,
         n_layers=num_layers,
-        hidden_size=config_data.hidden_size or 0,
-        # TODO: all custom models currently do not support tensor. We could add a dynamic test for this?
-        supports_tensor=model_card.metadata.supports_tensor
-        if model_card is not None
-        else False,
     )

src/exo/shared/tests/conftest.py

@@ -36,8 +36,6 @@ def get_pipeline_shard_metadata(
             pretty_name=str(model_id),
             storage_size=Memory.from_mb(100000),
             n_layers=32,
-            hidden_size=1000,
-            supports_tensor=True,
         ),
         device_rank=device_rank,
         world_size=world_size,

src/exo/shared/tests/test_apply/test_apply_node_download.py

@@ -19,7 +19,7 @@ def test_apply_node_download_progress():
         NodeDownloadProgress(download_progress=event), state
     )
 
-    assert new_state.downloads == {NodeId("node-1"): [event]}
+    assert new_state == State(downloads={NodeId("node-1"): [event]})
 
 
 def test_apply_two_node_download_progress():
@@ -42,4 +42,4 @@ def test_apply_two_node_download_progress():
     # TODO: This test is failing. We should support the following:
     # 1. Downloading multiple models concurrently on the same node (one per runner is fine).
     # 2. Downloading a model, it completes, then downloading a different model on the same node.
-    assert new_state.downloads == {NodeId("node-1"): [event1, event2]}
+    assert new_state == State(downloads={NodeId("node-1"): [event1, event2]})

src/exo/shared/types/api.py

@@ -5,7 +5,7 @@ from pydantic import BaseModel, Field, field_validator
 from pydantic_core import PydanticUseDefault
 
 from exo.shared.types.common import CommandId
-from exo.shared.types.models import ModelId, ModelMetadata
+from exo.shared.types.models import ModelId
 from exo.shared.types.worker.instances import Instance, InstanceId, InstanceMeta
 from exo.shared.types.worker.shards import Sharding
 
@@ -174,7 +174,6 @@ class DeleteInstanceTaskParams(BaseModel):
 class CreateInstanceResponse(BaseModel):
     message: str
     command_id: CommandId
-    model_meta: ModelMetadata
 
 
 class DeleteInstanceResponse(BaseModel):

src/exo/shared/types/models.py

@@ -14,5 +14,3 @@ class ModelMetadata(CamelCaseModel):
     pretty_name: str
     storage_size: Memory
     n_layers: PositiveInt
-    hidden_size: PositiveInt
-    supports_tensor: bool

src/exo/shared/types/tasks.py

@@ -40,10 +40,6 @@ class LoadModel(BaseTask):  # emitted by Worker
     pass
 
 
-class ConnectToGroup(BaseTask):  # emitted by Worker
-    pass
-
-
 class StartWarmup(BaseTask):  # emitted by Worker
     pass
 
@@ -61,11 +57,5 @@ class Shutdown(BaseTask):  # emitted by Worker
 
 
 Task = (
-    CreateRunner
-    | DownloadModel
-    | ConnectToGroup
-    | LoadModel
-    | StartWarmup
-    | ChatCompletion
-    | Shutdown
+    CreateRunner | DownloadModel | LoadModel | StartWarmup | ChatCompletion | Shutdown
 )

src/exo/shared/types/worker/runners.py

@@ -21,15 +21,7 @@ class BaseRunnerStatus(TaggedModel):
         return isinstance(self, RunnerRunning)
 
 
-class RunnerIdle(BaseRunnerStatus):
-    pass
-
-
-class RunnerConnecting(BaseRunnerStatus):
-    pass
-
-
-class RunnerConnected(BaseRunnerStatus):
+class RunnerWaitingForModel(BaseRunnerStatus):
     pass
 
 
@@ -62,9 +54,7 @@ class RunnerFailed(BaseRunnerStatus):
 
 
 RunnerStatus = (
-    RunnerIdle
-    | RunnerConnecting
-    | RunnerConnected
+    RunnerWaitingForModel
     | RunnerLoading
     | RunnerLoaded
     | RunnerWarmingUp

src/exo/worker/download/download_utils.py

@@ -450,11 +450,6 @@ async def get_weight_map(repo_id: str, revision: str = "main") -> dict[str, str]
 
 
 async def resolve_allow_patterns(shard: ShardMetadata) -> list[str]:
-    # TODO: 'Smart' downloads are disabled because:
-    #  (i) We don't handle all kinds of files;
-    # (ii) We don't have sticky sessions.
-    # (iii) Tensor parallel requires all files.
-    return ["*"]
     try:
         weight_map = await get_weight_map(str(shard.model_meta.model_id))
         return get_allow_patterns(weight_map, shard)

src/exo/worker/download/huggingface_utils.py

@@ -95,15 +95,7 @@ def extract_layer_num(tensor_name: str) -> int | None:
 
 def get_allow_patterns(weight_map: dict[str, str], shard: ShardMetadata) -> list[str]:
     default_patterns = set(
-        [
-            "*.json",
-            "*.py",
-            "tokenizer.model",
-            "tiktoken.model",
-            "*.tiktoken",
-            "*.txt",
-            "*.jinja",
-        ]
+        ["*.json", "*.py", "tokenizer.model", "*.tiktoken", "*.txt", "*.jinja"]
     )
     shard_specific_patterns: set[str] = set()
     if weight_map:

src/exo/worker/download/shard_downloader.py

@@ -1,5 +1,4 @@
 from abc import ABC, abstractmethod
-from copy import copy
 from datetime import timedelta
 from pathlib import Path
 from typing import AsyncIterator, Callable
@@ -13,7 +12,7 @@ from exo.shared.types.worker.shards import (
 from exo.worker.download.download_utils import RepoDownloadProgress
 
 
-# TODO: the PipelineShardMetadata getting reinstantiated is a bit messy. Should this be a classmethod?
+# TODO: the PipelineShardMetadata getting reinstantiated is a bit messy. Shoudl this be a classmethod?
 class ShardDownloader(ABC):
     @abstractmethod
     async def ensure_shard(
@@ -44,7 +43,34 @@ class ShardDownloader(ABC):
         Yields:
             tuple[Path, RepoDownloadProgress]: The path and progress of a shard download.
         """
-        yield (Path("/tmp/noop_shard"), NOOP_DOWNLOAD_PROGRESS)
+        yield (
+            Path("/tmp/noop_shard"),
+            RepoDownloadProgress(
+                repo_id="noop",
+                repo_revision="noop",
+                shard=PipelineShardMetadata(
+                    model_meta=ModelMetadata(
+                        model_id=ModelId("noop"),
+                        pretty_name="noope",
+                        storage_size=Memory.from_bytes(0),
+                        n_layers=1,
+                    ),
+                    device_rank=0,
+                    world_size=1,
+                    start_layer=0,
+                    end_layer=1,
+                    n_layers=1,
+                ),
+                completed_files=0,
+                total_files=0,
+                downloaded_bytes=Memory.from_bytes(0),
+                downloaded_bytes_this_session=Memory.from_bytes(0),
+                total_bytes=Memory.from_bytes(0),
+                overall_speed=0,
+                overall_eta=timedelta(seconds=0),
+                status="complete",
+            ),
+        )
 
     @abstractmethod
     async def get_shard_download_status_for_shard(
@@ -68,41 +94,46 @@ class NoopShardDownloader(ShardDownloader):
     ) -> AsyncIterator[tuple[Path, RepoDownloadProgress]]:
         yield (
             Path("/tmp/noop_shard"),
-            NOOP_DOWNLOAD_PROGRESS,
+            RepoDownloadProgress(
+                repo_id="noop",
+                repo_revision="noop",
+                shard=PipelineShardMetadata(
+                    model_meta=ModelMetadata(
+                        model_id=ModelId("noop"),
+                        pretty_name="noope",
+                        storage_size=Memory.from_bytes(0),
+                        n_layers=1,
+                    ),
+                    device_rank=0,
+                    world_size=1,
+                    start_layer=0,
+                    end_layer=1,
+                    n_layers=1,
+                ),
+                completed_files=0,
+                total_files=0,
+                downloaded_bytes=Memory.from_bytes(0),
+                downloaded_bytes_this_session=Memory.from_bytes(0),
+                total_bytes=Memory.from_bytes(0),
+                overall_speed=0,
+                overall_eta=timedelta(seconds=0),
+                status="complete",
+            ),
         )
 
     async def get_shard_download_status_for_shard(
         self, shard: ShardMetadata
     ) -> RepoDownloadProgress:
-        dp = copy(NOOP_DOWNLOAD_PROGRESS)
-        dp.shard = shard
-        return dp
-
-
-NOOP_DOWNLOAD_PROGRESS = RepoDownloadProgress(
-    repo_id="noop",
-    repo_revision="noop",
-    shard=PipelineShardMetadata(
-        model_meta=ModelMetadata(
-            model_id=ModelId("noop"),
-            pretty_name="noope",
-            storage_size=Memory.from_bytes(0),
-            n_layers=1,
-            hidden_size=1,
-            supports_tensor=False,
-        ),
-        device_rank=0,
-        world_size=1,
-        start_layer=0,
-        end_layer=1,
-        n_layers=1,
-    ),
-    completed_files=0,
-    total_files=0,
-    downloaded_bytes=Memory.from_bytes(0),
-    downloaded_bytes_this_session=Memory.from_bytes(0),
-    total_bytes=Memory.from_bytes(0),
-    overall_speed=0,
-    overall_eta=timedelta(seconds=0),
-    status="complete",
-)
+        return RepoDownloadProgress(
+            repo_id="noop",
+            repo_revision="noop",
+            shard=shard,
+            completed_files=0,
+            total_files=0,
+            downloaded_bytes=Memory.from_bytes(0),
+            downloaded_bytes_this_session=Memory.from_bytes(0),
+            total_bytes=Memory.from_bytes(0),
+            overall_speed=0,
+            overall_eta=timedelta(seconds=0),
+            status="complete",
+        )

src/exo/worker/engines/mlx/auto_parallel.py

@@ -22,6 +22,7 @@ from mlx_lm.models.qwen3_moe import Qwen3MoeSparseMoeBlock
 from exo.shared.types.worker.shards import (
     PipelineShardMetadata,
 )
+from exo.worker.runner.bootstrap import logger
 
 
 class _LayerCallable(Protocol):
@@ -170,6 +171,8 @@ def pipeline_auto_parallel(
 
     start_layer, end_layer = model_shard_meta.start_layer, model_shard_meta.end_layer
     device_rank, world_size = model_shard_meta.device_rank, model_shard_meta.world_size
+    logger.info(f"[RING3DBG] pipeline_auto_parallel: device_rank={device_rank} world_size={world_size}")
+    logger.info(f"[RING3DBG] layers: start={start_layer} end={end_layer} count={len(layers)}")
 
     layers = layers[start_layer:end_layer]
     layers[0] = PipelineFirstLayer(layers[0], device_rank, group=group)

src/exo/worker/engines/mlx/constants.py

@@ -9,7 +9,8 @@ MAX_KV_SIZE: int | None = 3200
 KEEP_KV_SIZE: int | None = 1600
 QUANTIZE_MODEL_MODE: str | None = "affine"
 CACHE_GROUP_SIZE: int = 64
-KV_CACHE_BITS: int | None = None
+KV_CACHE_BITS: int | None = 8
+TEMPERATURE: float = 1.0
 
 # TODO: We should really make this opt-in, but Kimi requires trust_remote_code=True
 TRUST_REMOTE_CODE: bool = True

src/exo/worker/engines/mlx/utils_mlx.py

@@ -13,6 +13,7 @@ from mlx_lm.tokenizer_utils import TokenizerWrapper
 from exo.worker.engines.mlx.constants import (
     CACHE_GROUP_SIZE,
     KV_CACHE_BITS,
+    TEMPERATURE,
     TRUST_REMOTE_CODE,
 )
 
@@ -20,8 +21,6 @@ try:
     from mlx_lm.tokenizer_utils import load_tokenizer
 except ImportError:
     from mlx_lm.tokenizer_utils import load as load_tokenizer  # type: ignore
-import contextlib
-
 import mlx.core as mx
 import mlx.nn as nn
 from mlx_lm.utils import load_model
@@ -49,7 +48,6 @@ from exo.worker.engines.mlx.auto_parallel import (
 )
 from exo.worker.runner.bootstrap import logger
 
-Group = mx.distributed.Group
 # Needed for 8 bit model
 resource.setrlimit(resource.RLIMIT_NOFILE, (2048, 4096))
 
@@ -69,7 +67,7 @@ def get_weights_size(model_shard_meta: ShardMetadata) -> Memory:
     )
 
 
-def mx_barrier(group: Group | None = None):
+def mx_barrier(group: mx.distributed.Group | None = None):
     mx.eval(
         mx.distributed.all_sum(
             mx.array(1.0),
@@ -79,7 +77,7 @@ def mx_barrier(group: Group | None = None):
     )
 
 
-def broadcast_from_zero(value: int, group: Group | None = None):
+def broadcast_from_zero(value: int, group: mx.distributed.Group | None = None):
     if group is None:
         return value
 
@@ -101,97 +99,97 @@ class HostList(RootModel[list[str]]):
 
 def mlx_distributed_init(
     bound_instance: BoundInstance,
-) -> Group:
+) -> mx.distributed.Group:
     """
-    Initialize MLX distributed.
+    Initialize the MLX distributed (runs in thread pool).
+
+    Either hosts or mlx_ibv_devices must be provided:
+    - hosts: traditional host-based connectivity using MLX_HOSTFILE
+    - mlx_ibv_devices: RDMA connectivity matrix using MLX_IBV_DEVICES
+    - mlx_ibv_coordinator: coordinator address (IP:PORT) for RDMA setup
+    - strict: if True, raise an error if the distributed backend is not available
     """
     rank = bound_instance.bound_shard.device_rank
     logger.info(f"Starting initialization for rank {rank}")
+    logger.info(f"[RING3DBG] mlx_distributed_init: bound_node_id={bound_instance.bound_node_id}")
+    logger.info(f"[RING3DBG] device_rank={rank}, world_size={bound_instance.bound_shard.world_size}")
 
-    coordination_file = None
-    try:
-        # TODO: singleton instances
-        match bound_instance.instance:
-            case MlxRingInstance(hosts_by_node=hosts_by_node, ephemeral_port=_):
-                coordination_file = (
-                    f"./hosts_{bound_instance.instance.instance_id}_{rank}.json"
-                )
-                hosts_for_node = hosts_by_node[bound_instance.bound_node_id]
-                hosts_json = HostList.from_hosts(hosts_for_node).model_dump_json()
+    # TODO: singleton instances
+    match bound_instance.instance:
+        case MlxRingInstance(hosts_by_node=hosts_by_node, ephemeral_port=_):
+            hostfile = f"./hosts_{rank}.json"
+            hosts_for_node = hosts_by_node[bound_instance.bound_node_id]
+            logger.info(f"[RING3DBG] hosts_by_node keys: {list(hosts_by_node.keys())}")
+            logger.info(f"[RING3DBG] hosts_for_node (len={len(hosts_for_node)}): {hosts_for_node}")
+            hosts_json = HostList.from_hosts(hosts_for_node).model_dump_json()
 
-                with open(coordination_file, "w") as f:
-                    _ = f.write(hosts_json)
+            with open(hostfile, "w") as f:
+                _ = f.write(hosts_json)
 
-                logger.info(
-                    f"rank {rank} hostfile: {coordination_file} hosts: {hosts_json}"
-                )
+            logger.info(f"rank {rank} hostfile: {hostfile} hosts: {hosts_json}")
 
-                os.environ["MLX_HOSTFILE"] = coordination_file
-                os.environ["MLX_RANK"] = str(rank)
-                os.environ["MLX_RING_VERBOSE"] = "1"
-                group = mx.distributed.init(backend="ring", strict=True)
+            os.environ["MLX_HOSTFILE"] = hostfile
+            os.environ["MLX_RANK"] = str(rank)
+            os.environ["MLX_RING_VERBOSE"] = "1"
+            group = mx.distributed.init(backend="ring", strict=True)
 
-            case MlxJacclInstance(
-                ibv_devices=ibv_devices, jaccl_coordinators=jaccl_coordinators
-            ):
-                # Use RDMA connectivity matrix
-                coordination_file = (
-                    f"./hosts_{bound_instance.instance.instance_id}_{rank}.json"
-                )
-                ibv_devices_json = json.dumps(ibv_devices)
+        case MlxJacclInstance(
+            ibv_devices=ibv_devices, jaccl_coordinators=jaccl_coordinators
+        ):
+            # Use RDMA connectivity matrix
+            devices_file = f"./hosts_{rank}.json"
+            ibv_devices_json = json.dumps(ibv_devices)
 
-                with open(coordination_file, "w") as f:
-                    _ = f.write(ibv_devices_json)
+            with open(devices_file, "w") as f:
+                _ = f.write(ibv_devices_json)
 
-                jaccl_coordinator = jaccl_coordinators[bound_instance.bound_node_id]
+            jaccl_coordinator = jaccl_coordinators[bound_instance.bound_node_id]
 
-                logger.info(f"rank {rank} MLX_IBV_DEVICES: {ibv_devices_json}")
-                logger.info(f"rank {rank} MLX_JACCL_COORDINATOR: {jaccl_coordinator}")
-                os.environ["MLX_IBV_DEVICES"] = coordination_file
-                os.environ["MLX_RANK"] = str(rank)
-                os.environ["MLX_JACCL_COORDINATOR"] = jaccl_coordinator
-                group = mx.distributed.init(backend="jaccl", strict=True)
+            logger.info(f"rank {rank} MLX_IBV_DEVICES: {ibv_devices_json}")
+            logger.info(f"rank {rank} MLX_JACCL_COORDINATOR: {jaccl_coordinator}")
+            os.environ["MLX_IBV_DEVICES"] = devices_file
+            os.environ["MLX_RANK"] = str(rank)
+            os.environ["MLX_JACCL_COORDINATOR"] = jaccl_coordinator
+            group = mx.distributed.init(backend="jaccl", strict=True)
 
-        logger.info(f"Rank {rank} mlx distributed initialization complete")
+    logger.info(f"Rank {rank} mlx distributed initialization complete")
+    logger.info(f"[RING3DBG] ring init complete: group.rank()={group.rank()} group.size()={group.size()}")
 
-        return group
-    finally:
-        with contextlib.suppress(FileNotFoundError):
-            if coordination_file:
-                os.remove(coordination_file)
+    return group
 
 
 def initialize_mlx(
     bound_instance: BoundInstance,
-) -> Group:
-    # should we unseed it?
-    # TODO: pass in seed from params
+) -> tuple[Model, TokenizerWrapper, Callable[[mx.array], mx.array]]:
+    """
+    Initialize the MLX model, tokenizer, and sampler. Runs in the MLX thread.
+    """
     mx.random.seed(42)
 
-    assert len(bound_instance.instance.shard_assignments.node_to_runner) > 1, (
-        "Tried to initialize mlx for a single node instance"
-    )
-    return mlx_distributed_init(bound_instance)
+    set_wired_limit_for_model(get_weights_size(bound_instance.bound_shard))
 
-
-def load_mlx_items(
-    bound_instance: BoundInstance, group: Group | None
-) -> tuple[Model, TokenizerWrapper, Callable[[mx.array], mx.array]]:
-    # TODO: pass temperature
-    sampler: Callable[[mx.array], mx.array] = make_sampler(temp=0.7)
+    sampler: Callable[[mx.array], mx.array] = make_sampler(temp=TEMPERATURE)
     logger.info("Created a sampler")
 
-    if group is None:
+    if len(bound_instance.instance.shard_assignments.node_to_runner) <= 1:
         logger.info(f"Single device used for {bound_instance.instance}")
         model_path = build_model_path(bound_instance.bound_shard.model_meta.model_id)
         start_time = time.perf_counter()
         model, _ = load_model(model_path, strict=True)
         end_time = time.perf_counter()
         logger.info(f"Time taken to load model: {(end_time - start_time):.2f}s")
+        if hasattr(model, "model") and isinstance(model.model, DeepseekV3Model):  # type: ignore
+            pass
+            # model, config = quantize_model(
+            #    model, config, group_size=KV_GROUP_SIZE, bits=ATTENTION_KV_BITS, quant_predicate=quant_predicate, mode=QUANTIZE_MODEL_MODE
+            # )
+
         tokenizer = get_tokenizer(model_path, bound_instance.bound_shard)
 
     else:
         logger.info("Starting distributed init")
+        group = mlx_distributed_init(bound_instance)
+
         start_time = time.perf_counter()
         model, tokenizer = shard_and_load(bound_instance.bound_shard, group=group)
         end_time = time.perf_counter()
@@ -201,12 +199,14 @@ def load_mlx_items(
 
     set_wired_limit_for_model(get_weights_size(bound_instance.bound_shard))
 
+    logger.debug(model)
+
     return cast(Model, model), tokenizer, sampler
 
 
 def shard_and_load(
     shard_metadata: ShardMetadata,
-    group: Group,
+    group: mx.distributed.Group,
 ) -> tuple[nn.Module, TokenizerWrapper]:
     model_path = build_model_path(shard_metadata.model_meta.model_id)
 
@@ -234,6 +234,8 @@ def shard_and_load(
     tokenizer = get_tokenizer(model_path, shard_metadata)
 
     logger.info(f"Group size: {group.size()}, group rank: {group.rank()}")
+    logger.info(f"[RING3DBG] shard_and_load: expected device_rank={shard_metadata.device_rank} world_size={shard_metadata.world_size}")
+    logger.info(f"[RING3DBG] actual group.rank()={group.rank()} group.size()={group.size()}")
 
     match shard_metadata:
         case TensorShardMetadata():
@@ -395,5 +397,11 @@ def set_wired_limit_for_model(model_size: Memory):
             "MB. This can be slow. See the documentation for possible work-arounds: "
             "https://github.com/ml-explore/mlx-lm/tree/main#large-models"
         )
+    kv_bytes = int(0.02 * model_bytes)
+    target_cache = int(1.10 * (model_bytes + kv_bytes))
+    target_cache = min(target_cache, max_rec_size)
+    mx.set_cache_limit(target_cache)
     mx.set_wired_limit(max_rec_size)
-    logger.info(f"Wired limit set to {max_rec_size}.")
+    logger.info(
+        f"Wired limit set to {max_rec_size}. Cache limit set to {target_cache}."
+    )

src/exo/worker/main.py

@@ -23,7 +23,6 @@ from exo.shared.types.events import (
     TopologyEdgeCreated,
     TopologyEdgeDeleted,
 )
-from exo.shared.types.models import ModelId
 from exo.shared.types.multiaddr import Multiaddr
 from exo.shared.types.profiling import MemoryPerformanceProfile, NodePerformanceProfile
 from exo.shared.types.state import State
@@ -84,7 +83,7 @@ class Worker:
         self.out_for_delivery: dict[EventId, ForwarderEvent] = {}
 
         self.state: State = State()
-        self.download_status: dict[ModelId, DownloadProgress] = {}
+        self.download_status: dict[ShardMetadata, DownloadProgress] = {}
         self.runners: dict[RunnerId, RunnerSupervisor] = {}
         self._tg: TaskGroup | None = None
 
@@ -129,7 +128,6 @@ class Worker:
             tg.start_soon(start_polling_node_metrics, resource_monitor_callback)
 
             tg.start_soon(start_polling_memory_metrics, memory_monitor_callback)
-            tg.start_soon(self._emit_existing_download_progress)
             tg.start_soon(self._connection_message_event_writer)
             tg.start_soon(self._resend_out_for_delivery)
             tg.start_soon(self._event_applier)
@@ -202,11 +200,11 @@ class Worker:
                         )
                     )
                 case DownloadModel(shard_metadata=shard):
-                    if shard.model_meta.model_id not in self.download_status:
+                    if shard not in self.download_status:
                         progress = DownloadPending(
                             shard_metadata=shard, node_id=self.node_id
                         )
-                        self.download_status[shard.model_meta.model_id] = progress
+                        self.download_status[shard] = progress
                         await self.event_sender.send(
                             NodeDownloadProgress(download_progress=progress)
                         )
@@ -219,7 +217,7 @@ class Worker:
                         progress = DownloadCompleted(
                             shard_metadata=shard, node_id=self.node_id
                         )
-                        self.download_status[shard.model_meta.model_id] = progress
+                        self.download_status[shard] = progress
                         await self.event_sender.send(
                             NodeDownloadProgress(download_progress=progress)
                         )
@@ -230,7 +228,7 @@ class Worker:
                             )
                         )
                     else:
-                        await self.event_sender.send(
+                        self.event_sender.send_nowait(
                             TaskStatusUpdated(
                                 task_id=task.task_id, task_status=TaskStatus.Running
                             )
@@ -351,7 +349,7 @@ class Worker:
                 initial_progress
             ),
         )
-        self.download_status[task.shard_metadata.model_meta.model_id] = status
+        self.download_status[task.shard_metadata] = status
         self.event_sender.send_nowait(NodeDownloadProgress(download_progress=status))
 
         last_progress_time = 0.0
@@ -365,7 +363,7 @@ class Worker:
             nonlocal last_progress_time
             if progress.status == "complete":
                 status = DownloadCompleted(shard_metadata=shard, node_id=self.node_id)
-                self.download_status[shard.model_meta.model_id] = status
+                self.download_status[shard] = status
                 # Footgun!
                 self.event_sender.send_nowait(
                     NodeDownloadProgress(download_progress=status)
@@ -386,7 +384,7 @@ class Worker:
                         progress
                     ),
                 )
-                self.download_status[shard.model_meta.model_id] = status
+                self.download_status[shard] = status
                 self.event_sender.send_nowait(
                     NodeDownloadProgress(download_progress=status)
                 )
@@ -446,40 +444,3 @@ class Worker:
                     await self.event_sender.send(TopologyEdgeDeleted(edge=conn))
 
             await anyio.sleep(10)
-
-    async def _emit_existing_download_progress(self) -> None:
-        try:
-            while True:
-                logger.info("Fetching and emitting existing download progress...")
-                async for (
-                    _,
-                    progress,
-                ) in self.shard_downloader.get_shard_download_status():
-                    if progress.status == "complete":
-                        status = DownloadCompleted(
-                            node_id=self.node_id, shard_metadata=progress.shard
-                        )
-                    elif progress.status in ["in_progress", "not_started"]:
-                        if progress.downloaded_bytes_this_session.in_bytes == 0:
-                            status = DownloadPending(
-                                node_id=self.node_id, shard_metadata=progress.shard
-                            )
-                        else:
-                            status = DownloadOngoing(
-                                node_id=self.node_id,
-                                shard_metadata=progress.shard,
-                                download_progress=map_repo_download_progress_to_download_progress_data(
-                                    progress
-                                ),
-                            )
-                    else:
-                        continue
-
-                    self.download_status[progress.shard.model_meta.model_id] = status
-                    await self.event_sender.send(
-                        NodeDownloadProgress(download_progress=status)
-                    )
-                logger.info("Done emitting existing download progress.")
-                await anyio.sleep(5 * 60)  # 5 minutes
-        except Exception as e:
-            logger.error(f"Error emitting existing download progress: {e}")

src/exo/worker/plan.py

@@ -3,10 +3,8 @@
 from collections.abc import Mapping, Sequence
 
 from exo.shared.types.common import NodeId
-from exo.shared.types.models import ModelId
 from exo.shared.types.tasks import (
     ChatCompletion,
-    ConnectToGroup,
     CreateRunner,
     DownloadModel,
     LoadModel,
@@ -16,25 +14,20 @@ from exo.shared.types.tasks import (
     TaskId,
     TaskStatus,
 )
-from exo.shared.types.worker.downloads import (
-    DownloadCompleted,
-    DownloadOngoing,
-    DownloadProgress,
-)
+from exo.shared.types.worker.downloads import DownloadCompleted, DownloadProgress
 from exo.shared.types.worker.instances import BoundInstance, Instance, InstanceId
 from exo.shared.types.worker.runners import (
-    RunnerConnected,
-    RunnerConnecting,
     RunnerFailed,
     RunnerId,
-    RunnerIdle,
     RunnerLoaded,
     RunnerLoading,
     RunnerReady,
     RunnerRunning,
     RunnerStatus,
+    RunnerWaitingForModel,
     RunnerWarmingUp,
 )
+from exo.shared.types.worker.shards import ShardMetadata
 from exo.worker.runner.runner_supervisor import RunnerSupervisor
 
 
@@ -43,7 +36,7 @@ def plan(
     # Runners is expected to be FRESH and so should not come from state
     runners: Mapping[RunnerId, RunnerSupervisor],
     # DL_status is expected to be FRESH and so should not come from state
-    download_status: Mapping[ModelId, DownloadProgress],
+    download_status: Mapping[ShardMetadata, DownloadProgress],
     # gdls is not expected to be fresh
     global_download_status: Mapping[NodeId, Sequence[DownloadProgress]],
     instances: Mapping[InstanceId, Instance],
@@ -55,7 +48,6 @@ def plan(
         _kill_runner(runners, all_runners, instances)
         or _create_runner(node_id, runners, instances)
         or _model_needs_download(runners, download_status)
-        or _init_distributed_backend(runners, all_runners)
         or _load_model(runners, all_runners, global_download_status)
         or _ready_to_warmup(runners, all_runners)
         or _pending_tasks(runners, tasks, all_runners)
@@ -111,15 +103,12 @@ def _create_runner(
 
 def _model_needs_download(
     runners: Mapping[RunnerId, RunnerSupervisor],
-    download_status: Mapping[ModelId, DownloadProgress],
+    download_status: Mapping[ShardMetadata, DownloadProgress],
 ) -> DownloadModel | None:
     for runner in runners.values():
-        model_id = runner.bound_instance.bound_shard.model_meta.model_id
-        if isinstance(runner.status, RunnerIdle) and (
-            model_id not in download_status
-            or not isinstance(
-                download_status[model_id], (DownloadOngoing, DownloadCompleted)
-            )
+        if (
+            isinstance(runner.status, RunnerWaitingForModel)
+            and runner.bound_instance.bound_shard not in download_status
         ):
             # We don't invalidate download_status randomly in case a file gets deleted on disk
             return DownloadModel(
@@ -128,54 +117,14 @@ def _model_needs_download(
             )
 
 
-def _init_distributed_backend(
+""" --- TODO!
+def _init_backend(
     runners: Mapping[RunnerId, RunnerSupervisor],
     all_runners: Mapping[RunnerId, RunnerStatus],
-):
-    for runner in runners.values():
-        instance = runner.bound_instance.instance
-        shard_assignments = instance.shard_assignments
-
-        is_single_node_instance = len(shard_assignments.runner_to_shard) == 1
-        if is_single_node_instance:
-            continue
-
-        runner_is_idle = isinstance(runner.status, RunnerIdle)
-        all_runners_connecting = all(
-            isinstance(
-                all_runners.get(global_runner_id),
-                (RunnerConnecting, RunnerIdle),
-            )
-            for global_runner_id in shard_assignments.runner_to_shard
-        )
-
-        if not (runner_is_idle and all_runners_connecting):
-            continue
-
-        runner_id = runner.bound_instance.bound_runner_id
-
-        shard = runner.bound_instance.bound_shard
-        device_rank = shard.device_rank
-        world_size = shard.world_size
-
-        assert device_rank < world_size
-        assert device_rank >= 0
-
-        accepting_ranks = device_rank < world_size - 1
-
-        # Rank = n-1
-        connecting_rank_ready = device_rank == world_size - 1 and all(
-            isinstance(all_runners.get(global_runner_id, None), RunnerConnecting)
-            for global_runner_id in shard_assignments.runner_to_shard
-            if global_runner_id != runner_id
-        )
-
-        if not (accepting_ranks or connecting_rank_ready):
-            continue
-
-        return ConnectToGroup(instance_id=instance.instance_id)
-
-    return None
+) -> LoadModel | None:
+    for runner in runner.values()
+    pass
+"""
 
 
 def _load_model(
@@ -187,33 +136,31 @@ def _load_model(
         instance = runner.bound_instance.instance
         shard_assignments = instance.shard_assignments
 
-        all_local_downloads_complete = all(
+        all_downloads_complete_local = all(
             nid in global_download_status
             and any(
                 isinstance(dp, DownloadCompleted)
-                and dp.shard_metadata.model_meta.model_id == shard_assignments.model_id
+                and dp.shard_metadata == shard_assignments.runner_to_shard[rid]
                 for dp in global_download_status[nid]
             )
-            for nid in shard_assignments.node_to_runner
+            for nid, rid in shard_assignments.node_to_runner.items()
         )
-        if not all_local_downloads_complete:
-            continue
 
-        is_single_node_instance = len(instance.shard_assignments.runner_to_shard) == 1
-        if is_single_node_instance and isinstance(runner.status, RunnerIdle):
-            return LoadModel(instance_id=instance.instance_id)
+        runner_is_waiting = isinstance(runner.status, RunnerWaitingForModel)
 
-        is_runner_waiting = isinstance(runner.status, RunnerConnected)
-
-        all_ready_for_model = all(
+        all_runners_expecting_model = all(
             isinstance(
-                all_runners.get(global_runner_id, None),
-                (RunnerConnected, RunnerLoading, RunnerLoaded),
+                all_runners.get(global_runner_id),
+                (RunnerWaitingForModel, RunnerLoading, RunnerLoaded),
             )
             for global_runner_id in shard_assignments.runner_to_shard
         )
 
-        if is_runner_waiting and all_ready_for_model:
+        if (
+            all_downloads_complete_local
+            and runner_is_waiting
+            and all_runners_expecting_model
+        ):
             return LoadModel(instance_id=instance.instance_id)
 
     return None
@@ -236,8 +183,8 @@ def _ready_to_warmup(
         assert device_rank < world_size
         assert device_rank >= 0
 
-        # Rank != 0
-        accepting_ranks_ready = device_rank > 0 and all(
+        # Rank != n-1
+        accepting_ranks_ready = device_rank != world_size - 1 and all(
             isinstance(
                 all_runners.get(global_runner_id, None),
                 (RunnerLoaded, RunnerWarmingUp),
@@ -245,8 +192,8 @@ def _ready_to_warmup(
             for global_runner_id in shard_assignments.runner_to_shard
         )
 
-        # Rank = 0
-        connecting_rank_ready = device_rank == 0 and all(
+        # Rank = n-1
+        connecting_rank_ready = device_rank == world_size - 1 and all(
             isinstance(all_runners.get(global_runner_id, None), RunnerWarmingUp)
             for global_runner_id in shard_assignments.runner_to_shard
             if global_runner_id != runner_id
@@ -274,8 +221,6 @@ def _pending_tasks(
             if task.instance_id != runner.bound_instance.instance.instance_id:
                 continue
 
-            # TODO: Check ordering aligns with MLX distributeds expectations.
-
             if isinstance(runner.status, RunnerReady) and all(
                 isinstance(all_runners[global_runner_id], (RunnerReady, RunnerRunning))
                 for global_runner_id in runner.bound_instance.instance.shard_assignments.runner_to_shard

src/exo/worker/runner/bootstrap.py

@@ -2,13 +2,16 @@ import os
 
 import loguru
 
-from exo.shared.types.events import Event, RunnerStatusUpdated
+from exo.shared.types.events import Event
 from exo.shared.types.tasks import Task
 from exo.shared.types.worker.instances import BoundInstance, MlxJacclInstance
-from exo.shared.types.worker.runners import RunnerFailed
 from exo.utils.channels import MpReceiver, MpSender
 
-logger: "loguru.Logger" = loguru.logger
+logger: "loguru.Logger"
+
+
+if os.getenv("EXO_TESTS") == "1":
+    logger = loguru.logger
 
 
 def entrypoint(
@@ -27,23 +30,6 @@ def entrypoint(
     logger = _logger
 
     # Import main after setting global logger - this lets us just import logger from this module
-    try:
-        from exo.worker.runner.runner import main
+    from exo.worker.runner.runner import main
 
-        main(bound_instance, event_sender, task_receiver)
-    except Exception as e:
-        logger.opt(exception=e).warning(
-            f"Runner {bound_instance.bound_runner_id} crashed with critical exception {e}"
-        )
-        event_sender.send(
-            RunnerStatusUpdated(
-                runner_id=bound_instance.bound_runner_id,
-                runner_status=RunnerFailed(error_message=str(e)),
-            )
-        )
-    finally:
-        event_sender.close()
-        task_receiver.close()
-        event_sender.join()
-        task_receiver.join()
-        logger.info("bye from the runner")
+    main(bound_instance, event_sender, task_receiver)

src/exo/worker/runner/runner.py

@@ -11,7 +11,6 @@ from exo.shared.types.events import (
 )
 from exo.shared.types.tasks import (
     ChatCompletion,
-    ConnectToGroup,
     LoadModel,
     Shutdown,
     StartWarmup,
@@ -23,23 +22,20 @@ from exo.shared.types.worker.runner_response import (
     GenerationResponse,
 )
 from exo.shared.types.worker.runners import (
-    RunnerConnected,
-    RunnerConnecting,
     RunnerFailed,
-    RunnerIdle,
     RunnerLoaded,
     RunnerLoading,
     RunnerReady,
     RunnerRunning,
     RunnerShutdown,
     RunnerStatus,
+    RunnerWaitingForModel,
     RunnerWarmingUp,
 )
 from exo.utils.channels import ClosedResourceError, MpReceiver, MpSender
 from exo.worker.engines.mlx.generator.generate import mlx_generate, warmup_inference
 from exo.worker.engines.mlx.utils_mlx import (
     initialize_mlx,
-    load_mlx_items,
     mlx_force_oom,
 )
 from exo.worker.runner.bootstrap import logger
@@ -67,10 +63,9 @@ def main(
         model = None
         tokenizer = None
         sampler = None
-        group = None
 
-        current_status: RunnerStatus = RunnerIdle()
-        logger.info("runner created")
+        current_status: RunnerStatus = RunnerWaitingForModel()
+        logger.info("runner waiting for model")
         event_sender.send(
             RunnerStatusUpdated(runner_id=runner_id, runner_status=current_status)
         )
@@ -83,26 +78,9 @@ def main(
                 )
                 event_sender.send(TaskAcknowledged(task_id=task.task_id))
                 match task:
-                    case ConnectToGroup() if isinstance(
-                        current_status, (RunnerIdle, RunnerFailed)
+                    case LoadModel() if isinstance(
+                        current_status, (RunnerWaitingForModel, RunnerFailed)
                     ):
-                        logger.info("runner connecting")
-                        current_status = RunnerConnecting()
-                        event_sender.send(
-                            RunnerStatusUpdated(
-                                runner_id=runner_id, runner_status=current_status
-                            )
-                        )
-                        group = initialize_mlx(bound_instance)
-
-                        logger.info("runner connected")
-                        current_status = RunnerConnected()
-
-                    # we load the model if it's connected with a group, or idle without a group. we should never tell a model to connect if it doesn't need to
-                    case LoadModel() if (
-                        isinstance(current_status, RunnerConnected)
-                        and group is not None
-                    ) or (isinstance(current_status, RunnerIdle) and group is None):
                         current_status = RunnerLoading()
                         logger.info("runner loading")
                         event_sender.send(
@@ -111,12 +89,15 @@ def main(
                             )
                         )
 
-                        model, tokenizer, sampler = load_mlx_items(
-                            bound_instance, group
-                        )
+                        model, tokenizer, sampler = initialize_mlx(bound_instance)
 
                         current_status = RunnerLoaded()
                         logger.info("runner loaded")
+                        event_sender.send(
+                            RunnerStatusUpdated(
+                                runner_id=runner_id, runner_status=current_status
+                            )
+                        )
                     case StartWarmup() if isinstance(current_status, RunnerLoaded):
                         assert model
                         assert tokenizer
@@ -142,6 +123,11 @@ def main(
                         )
                         current_status = RunnerReady()
                         logger.info("runner ready")
+                        event_sender.send(
+                            RunnerStatusUpdated(
+                                runner_id=runner_id, runner_status=RunnerReady()
+                            )
+                        )
                     case ChatCompletion(
                         task_params=task_params, command_id=command_id
                     ) if isinstance(current_status, RunnerReady):
@@ -186,6 +172,11 @@ def main(
 
                         current_status = RunnerReady()
                         logger.info("runner ready")
+                        event_sender.send(
+                            RunnerStatusUpdated(
+                                runner_id=runner_id, runner_status=RunnerReady()
+                            )
+                        )
                     case Shutdown():
                         logger.info("runner shutting down")
                         event_sender.send(
@@ -195,19 +186,12 @@ def main(
                         )
                         break
                     case _:
-                        raise ValueError(
-                            f"Received {task.__class__.__name__} outside of state machine in {current_status=}"
-                        )
+                        raise ValueError("Received task outside of state machine")
                 event_sender.send(
                     TaskStatusUpdated(
                         task_id=task.task_id, task_status=TaskStatus.Complete
                     )
                 )
-                event_sender.send(
-                    RunnerStatusUpdated(
-                        runner_id=runner_id, runner_status=current_status
-                    )
-                )
         event_sender.send(
             RunnerStatusUpdated(runner_id=runner_id, runner_status=RunnerShutdown())
         )

src/exo/worker/runner/runner_supervisor.py

@@ -19,8 +19,8 @@ from exo.shared.types.tasks import Task, TaskId
 from exo.shared.types.worker.instances import BoundInstance
 from exo.shared.types.worker.runners import (
     RunnerFailed,
-    RunnerIdle,
     RunnerStatus,
+    RunnerWaitingForModel,
 )
 from exo.shared.types.worker.shards import ShardMetadata
 from exo.utils.channels import MpReceiver, MpSender, Sender, mp_channel
@@ -41,7 +41,7 @@ class RunnerSupervisor:
     _event_sender: Sender[Event]
     # err_path: str
     _tg: TaskGroup | None = field(default=None, init=False)
-    status: RunnerStatus = field(default_factory=RunnerIdle, init=False)
+    status: RunnerStatus = field(default_factory=RunnerWaitingForModel, init=False)
     pending: dict[TaskId, anyio.Event] = field(default_factory=dict, init=False)
 
     @classmethod

src/exo/worker/tests/constants.py

@@ -9,11 +9,9 @@ MASTER_NODE_ID = NodeId("ffffffff-aaaa-4aaa-8aaa-aaaaaaaaaaaa")
 
 NODE_A: Final[NodeId] = NodeId("aaaaaaaa-aaaa-4aaa-8aaa-aaaaaaaaaaaa")
 NODE_B: Final[NodeId] = NodeId("bbbbbbbb-bbbb-4bbb-8bbb-bbbbbbbbbbbb")
-NODE_C: Final[NodeId] = NodeId("cccccccc-cccc-4ccc-8ccc-cccccccccccc")
 
 RUNNER_1_ID: Final[RunnerId] = RunnerId("11111111-1111-4111-8111-111111111111")
 RUNNER_2_ID: Final[RunnerId] = RunnerId("33333333-3333-4333-8333-333333333333")
-RUNNER_3_ID: Final[RunnerId] = RunnerId("Runner3")
 
 INSTANCE_1_ID: Final[InstanceId] = InstanceId("22222222-2222-4222-8222-222222222222")
 INSTANCE_2_ID: Final[InstanceId] = InstanceId("44444444-4444-4444-8444-444444444444")
@@ -26,9 +24,3 @@ TASK_2_ID: Final[TaskId] = TaskId("66666666-6666-4666-8666-666666666666")
 
 COMMAND_1_ID: Final[CommandId] = CommandId("77777777-7777-4777-8777-777777777777")
 COMMAND_2_ID: Final[CommandId] = CommandId("88888888-8888-4888-8888-888888888888")
-
-SHUTDOWN_TASK_ID = TaskId("shutdown")
-CHAT_COMPLETION_TASK_ID = TaskId("chat-completion")
-INITIALIZATION_TASK_ID = TaskId("initialisation")
-LOAD_TASK_ID = TaskId("load")
-WARMUP_TASK_ID = TaskId("warmup")

src/exo/worker/tests/unittests/conftest.py

@@ -1,5 +1,3 @@
-from __future__ import annotations
-
 from dataclasses import dataclass
 
 from exo.shared.types.common import NodeId
@@ -16,7 +14,6 @@ from exo.shared.types.worker.runners import RunnerId, RunnerStatus, ShardAssignm
 from exo.shared.types.worker.shards import PipelineShardMetadata, ShardMetadata
 
 
-# Runner supervisor without multiprocessing logic.
 @dataclass(frozen=True)
 class FakeRunnerSupervisor:
     bound_instance: BoundInstance
@@ -38,8 +35,6 @@ def get_pipeline_shard_metadata(
             pretty_name=str(model_id),
             storage_size=Memory.from_mb(100000),
             n_layers=32,
-            hidden_size=2048,
-            supports_tensor=False,
         ),
         device_rank=device_rank,
         world_size=world_size,
@@ -75,24 +70,3 @@ def get_mlx_ring_instance(
         hosts_by_node={},
         ephemeral_port=50000,
     )
-
-
-def get_bound_mlx_ring_instance(
-    instance_id: InstanceId, model_id: ModelId, runner_id: RunnerId, node_id: NodeId
-) -> BoundInstance:
-    shard = get_pipeline_shard_metadata(model_id=model_id, device_rank=0, world_size=2)
-    other_shard = get_pipeline_shard_metadata(
-        model_id=model_id, device_rank=1, world_size=2
-    )
-    instance = get_mlx_ring_instance(
-        instance_id=instance_id,
-        model_id=model_id,
-        node_to_runner={
-            node_id: runner_id,
-            NodeId("other_node"): RunnerId("other_runner"),
-        },
-        runner_to_shard={runner_id: shard, RunnerId("other_runner"): other_shard},
-    )
-    return BoundInstance(
-        instance=instance, bound_runner_id=runner_id, bound_node_id=node_id
-    )

src/exo/worker/tests/unittests/test_plan/test_download_and_loading.py

@@ -1,13 +1,12 @@
 import exo.worker.plan as plan_mod
 from exo.shared.types.common import NodeId
-from exo.shared.types.models import ModelId
 from exo.shared.types.tasks import LoadModel
 from exo.shared.types.worker.downloads import DownloadCompleted, DownloadProgress
 from exo.shared.types.worker.instances import BoundInstance
 from exo.shared.types.worker.runners import (
-    RunnerConnected,
-    RunnerIdle,
+    RunnerWaitingForModel,
 )
+from exo.shared.types.worker.shards import ShardMetadata
 from exo.worker.tests.constants import (
     INSTANCE_1_ID,
     MODEL_A_ID,
@@ -39,14 +38,16 @@ def test_plan_requests_download_when_waiting_and_shard_not_downloaded():
     bound_instance = BoundInstance(
         instance=instance, bound_runner_id=RUNNER_1_ID, bound_node_id=NODE_A
     )
-    runner = FakeRunnerSupervisor(bound_instance=bound_instance, status=RunnerIdle())
+    runner = FakeRunnerSupervisor(
+        bound_instance=bound_instance, status=RunnerWaitingForModel()
+    )
 
     runners = {RUNNER_1_ID: runner}
     instances = {INSTANCE_1_ID: instance}
-    all_runners = {RUNNER_1_ID: RunnerIdle()}
+    all_runners = {RUNNER_1_ID: RunnerWaitingForModel()}
 
     # No entry for this shard -> should trigger DownloadModel
-    download_status: dict[ModelId, DownloadProgress] = {}
+    download_status: dict[ShardMetadata, DownloadProgress] = {}
 
     result = plan_mod.plan(
         node_id=NODE_A,
@@ -81,20 +82,20 @@ def test_plan_loads_model_when_all_shards_downloaded_and_waiting():
         instance=instance, bound_runner_id=RUNNER_1_ID, bound_node_id=NODE_A
     )
     local_runner = FakeRunnerSupervisor(
-        bound_instance=bound_instance, status=RunnerConnected()
+        bound_instance=bound_instance, status=RunnerWaitingForModel()
     )
 
     runners = {RUNNER_1_ID: local_runner}
     instances = {INSTANCE_1_ID: instance}
 
     all_runners = {
-        RUNNER_1_ID: RunnerConnected(),
-        RUNNER_2_ID: RunnerConnected(),
+        RUNNER_1_ID: RunnerWaitingForModel(),
+        RUNNER_2_ID: RunnerWaitingForModel(),
     }
 
     # Local node has already marked its shard as downloaded (not actually used by _load_model)
     local_download_status = {
-        MODEL_A_ID: DownloadCompleted(shard_metadata=shard1, node_id=NODE_A)
+        shard1: DownloadCompleted(shard_metadata=shard1, node_id=NODE_A)  # type: ignore[reportUnhashable]
     }
 
     # Global view has completed downloads for both nodes
@@ -132,15 +133,17 @@ def test_plan_does_not_request_download_when_shard_already_downloaded():
     bound_instance = BoundInstance(
         instance=instance, bound_runner_id=RUNNER_1_ID, bound_node_id=NODE_A
     )
-    runner = FakeRunnerSupervisor(bound_instance=bound_instance, status=RunnerIdle())
+    runner = FakeRunnerSupervisor(
+        bound_instance=bound_instance, status=RunnerWaitingForModel()
+    )
 
     runners = {RUNNER_1_ID: runner}
     instances = {INSTANCE_1_ID: instance}
-    all_runners = {RUNNER_1_ID: RunnerIdle()}
+    all_runners = {RUNNER_1_ID: RunnerWaitingForModel()}
 
     # Local status claims the shard is downloaded already
     local_download_status = {
-        MODEL_A_ID: DownloadCompleted(shard_metadata=shard, node_id=NODE_A)
+        shard: DownloadCompleted(shard_metadata=shard, node_id=NODE_A)  # type: ignore[reportUnhashable]
     }
 
     # Global view hasn't caught up yet (no completed shards recorded for NODE_A)
@@ -180,19 +183,19 @@ def test_plan_does_not_load_model_until_all_shards_downloaded_globally():
         instance=instance, bound_runner_id=RUNNER_1_ID, bound_node_id=NODE_A
     )
     local_runner = FakeRunnerSupervisor(
-        bound_instance=bound_instance, status=RunnerConnected()
+        bound_instance=bound_instance, status=RunnerWaitingForModel()
     )
 
     runners = {RUNNER_1_ID: local_runner}
     instances = {INSTANCE_1_ID: instance}
     all_runners = {
-        RUNNER_1_ID: RunnerConnected(),
-        RUNNER_2_ID: RunnerConnected(),
+        RUNNER_1_ID: RunnerWaitingForModel(),
+        RUNNER_2_ID: RunnerWaitingForModel(),
     }
 
     # Only NODE_A's shard is recorded as downloaded globally
     local_download_status = {
-        MODEL_A_ID: DownloadCompleted(shard_metadata=shard1, node_id=NODE_A)
+        shard1: DownloadCompleted(shard_metadata=shard1, node_id=NODE_A)  # type: ignore[reportUnhashable]
     }
     global_download_status = {
         NODE_A: [DownloadCompleted(shard_metadata=shard1, node_id=NODE_A)],
@@ -210,22 +213,3 @@ def test_plan_does_not_load_model_until_all_shards_downloaded_globally():
     )
 
     assert result is None
-
-    global_download_status = {
-        NODE_A: [DownloadCompleted(shard_metadata=shard1, node_id=NODE_A)],
-        NODE_B: [
-            DownloadCompleted(shard_metadata=shard2, node_id=NODE_B)
-        ],  # NODE_B has no downloads completed yet
-    }
-
-    result = plan_mod.plan(
-        node_id=NODE_A,
-        runners=runners,  # type: ignore
-        download_status=local_download_status,
-        global_download_status=global_download_status,
-        instances=instances,
-        all_runners=all_runners,
-        tasks={},
-    )
-
-    assert result is not None

src/exo/worker/tests/unittests/test_plan/test_task_forwarding.py

@@ -5,9 +5,9 @@ from exo.shared.types.api import ChatCompletionTaskParams
 from exo.shared.types.tasks import ChatCompletion, Task, TaskId, TaskStatus
 from exo.shared.types.worker.instances import BoundInstance, InstanceId
 from exo.shared.types.worker.runners import (
-    RunnerIdle,
     RunnerReady,
     RunnerRunning,
+    RunnerWaitingForModel,
 )
 from exo.worker.tests.constants import (
     COMMAND_1_ID,
@@ -99,7 +99,7 @@ def test_plan_does_not_forward_chat_completion_if_any_runner_not_ready():
     instances = {INSTANCE_1_ID: instance}
     all_runners = {
         RUNNER_1_ID: RunnerReady(),
-        RUNNER_2_ID: RunnerIdle(),
+        RUNNER_2_ID: RunnerWaitingForModel(),
     }
 
     task = ChatCompletion(

src/exo/worker/tests/unittests/test_plan/test_warmup.py

@@ -2,9 +2,8 @@ import exo.worker.plan as plan_mod
 from exo.shared.types.tasks import StartWarmup
 from exo.shared.types.worker.instances import BoundInstance
 from exo.shared.types.worker.runners import (
-    RunnerIdle,
     RunnerLoaded,
-    RunnerLoading,
+    RunnerWaitingForModel,
     RunnerWarmingUp,
 )
 from exo.worker.tests.constants import (
@@ -12,10 +11,8 @@ from exo.worker.tests.constants import (
     MODEL_A_ID,
     NODE_A,
     NODE_B,
-    NODE_C,
     RUNNER_1_ID,
     RUNNER_2_ID,
-    RUNNER_3_ID,
 )
 from exo.worker.tests.unittests.conftest import (
     FakeRunnerSupervisor,
@@ -26,137 +23,9 @@ from exo.worker.tests.unittests.conftest import (
 
 def test_plan_starts_warmup_for_accepting_rank_when_all_loaded_or_warming():
     """
-    For non-zero device_rank shards, StartWarmup should be emitted when all
-    shards in the instance are Loaded/WarmingUp.
-    """
-    shard0 = get_pipeline_shard_metadata(MODEL_A_ID, device_rank=0, world_size=3)
-    shard1 = get_pipeline_shard_metadata(MODEL_A_ID, device_rank=1, world_size=3)
-    shard2 = get_pipeline_shard_metadata(MODEL_A_ID, device_rank=2, world_size=3)
-    instance = get_mlx_ring_instance(
-        instance_id=INSTANCE_1_ID,
-        model_id=MODEL_A_ID,
-        node_to_runner={NODE_A: RUNNER_1_ID, NODE_B: RUNNER_2_ID, NODE_C: RUNNER_3_ID},
-        runner_to_shard={RUNNER_1_ID: shard0, RUNNER_2_ID: shard1, RUNNER_3_ID: shard2},
-    )
-
-    bound_instance = BoundInstance(
-        instance=instance, bound_runner_id=RUNNER_2_ID, bound_node_id=NODE_B
-    )
-    local_runner = FakeRunnerSupervisor(
-        bound_instance=bound_instance, status=RunnerLoaded()
-    )
-
-    runners = {RUNNER_2_ID: local_runner}
-    instances = {INSTANCE_1_ID: instance}
-    all_runners = {
-        RUNNER_1_ID: RunnerLoaded(),
-        RUNNER_2_ID: RunnerLoaded(),
-        RUNNER_3_ID: RunnerWarmingUp(),
-    }
-
-    result = plan_mod.plan(
-        node_id=NODE_B,
-        runners=runners,  # type: ignore
-        download_status={},
-        global_download_status={NODE_A: []},
-        instances=instances,
-        all_runners=all_runners,
-        tasks={},
-    )
-
-    assert isinstance(result, StartWarmup)
-    assert result.instance_id == INSTANCE_1_ID
-
-
-def test_plan_starts_warmup_for_rank_zero_after_others_warming():
-    """
-    For device_rank == 0, StartWarmup should only be emitted once all the
-    other runners in the instance are already warming up.
-    """
-    shard0 = get_pipeline_shard_metadata(MODEL_A_ID, device_rank=0, world_size=2)
-    shard1 = get_pipeline_shard_metadata(MODEL_A_ID, device_rank=1, world_size=2)
-    instance = get_mlx_ring_instance(
-        instance_id=INSTANCE_1_ID,
-        model_id=MODEL_A_ID,
-        node_to_runner={NODE_A: RUNNER_1_ID, NODE_B: RUNNER_2_ID},
-        runner_to_shard={RUNNER_1_ID: shard0, RUNNER_2_ID: shard1},
-    )
-
-    bound_instance = BoundInstance(
-        instance=instance, bound_runner_id=RUNNER_1_ID, bound_node_id=NODE_A
-    )
-    local_runner = FakeRunnerSupervisor(
-        bound_instance=bound_instance, status=RunnerLoaded()
-    )
-
-    runners = {RUNNER_1_ID: local_runner}
-    instances = {INSTANCE_1_ID: instance}
-    all_runners = {
-        RUNNER_1_ID: RunnerLoaded(),
-        RUNNER_2_ID: RunnerWarmingUp(),
-    }
-
-    result = plan_mod.plan(
-        node_id=NODE_A,
-        runners=runners,  # type: ignore
-        download_status={},
-        global_download_status={NODE_A: []},
-        instances=instances,
-        all_runners=all_runners,
-        tasks={},
-    )
-
-    assert isinstance(result, StartWarmup)
-    assert result.instance_id == INSTANCE_1_ID
-
-
-def test_plan_does_not_start_warmup_for_non_zero_rank_until_all_loaded_or_warming():
-    """
-    Non-zero rank should not start warmup while any shard is not Loaded/WarmingUp.
-    """
-    shard0 = get_pipeline_shard_metadata(MODEL_A_ID, device_rank=0, world_size=2)
-    shard1 = get_pipeline_shard_metadata(MODEL_A_ID, device_rank=1, world_size=2)
-    instance = get_mlx_ring_instance(
-        instance_id=INSTANCE_1_ID,
-        model_id=MODEL_A_ID,
-        node_to_runner={NODE_A: RUNNER_1_ID, NODE_B: RUNNER_2_ID},
-        runner_to_shard={RUNNER_1_ID: shard0, RUNNER_2_ID: shard1},
-    )
-
-    bound_instance = BoundInstance(
-        instance=instance, bound_runner_id=RUNNER_2_ID, bound_node_id=NODE_B
-    )
-    local_runner = FakeRunnerSupervisor(
-        bound_instance=bound_instance, status=RunnerLoaded()
-    )
-
-    runners = {RUNNER_2_ID: local_runner}
-    instances = {INSTANCE_1_ID: instance}
-    all_runners = {
-        RUNNER_1_ID: RunnerIdle(),
-        RUNNER_2_ID: RunnerLoaded(),
-    }
-
-    result = plan_mod.plan(
-        node_id=NODE_B,
-        runners=runners,  # type: ignore
-        download_status={},
-        global_download_status={NODE_A: [], NODE_B: []},
-        instances=instances,
-        all_runners=all_runners,
-        tasks={},
-    )
-
-    assert result is None
-
-
-def test_plan_does_not_start_warmup_for_rank_zero_until_others_warming():
-    """
-    Rank-zero shard should not start warmup until all non-zero ranks are
-    already WarmingUp.
-    For accepting ranks (device_rank != 0), StartWarmup should be
+    For accepting ranks (device_rank != world_size - 1), StartWarmup should be
     emitted when all shards in the instance are Loaded/WarmingUp.
-    In a 2-node setup, rank 1 is the accepting rank.
+    In a 2-node setup, rank 0 is the accepting rank.
     """
     shard0 = get_pipeline_shard_metadata(MODEL_A_ID, device_rank=0, world_size=2)
     shard1 = get_pipeline_shard_metadata(MODEL_A_ID, device_rank=1, world_size=2)
@@ -167,7 +36,7 @@ def test_plan_does_not_start_warmup_for_rank_zero_until_others_warming():
         runner_to_shard={RUNNER_1_ID: shard0, RUNNER_2_ID: shard1},
     )
 
-    # Rank 1 is the accepting rank
+    # Rank 0 is the accepting rank
     bound_instance = BoundInstance(
         instance=instance, bound_runner_id=RUNNER_1_ID, bound_node_id=NODE_A
     )
@@ -192,23 +61,6 @@ def test_plan_does_not_start_warmup_for_rank_zero_until_others_warming():
         tasks={},
     )
 
-    assert result is None
-
-    all_runners = {
-        RUNNER_1_ID: RunnerLoaded(),
-        RUNNER_2_ID: RunnerWarmingUp(),
-    }
-
-    result = plan_mod.plan(
-        node_id=NODE_A,
-        runners=runners,  # type: ignore
-        download_status={},
-        global_download_status={NODE_A: []},
-        instances=instances,
-        all_runners=all_runners,
-        tasks={},
-    )
-
     assert isinstance(result, StartWarmup)
     assert result.instance_id == INSTANCE_1_ID
 
@@ -283,7 +135,7 @@ def test_plan_does_not_start_warmup_for_accepting_rank_until_all_loaded_or_warmi
     instances = {INSTANCE_1_ID: instance}
     all_runners = {
         RUNNER_1_ID: RunnerLoaded(),
-        RUNNER_2_ID: RunnerLoading(),
+        RUNNER_2_ID: RunnerWaitingForModel(),
     }
 
     result = plan_mod.plan(
@@ -301,8 +153,9 @@ def test_plan_does_not_start_warmup_for_accepting_rank_until_all_loaded_or_warmi
 
 def test_plan_does_not_start_warmup_for_connecting_rank_until_others_warming():
     """
-    Connecting rank (device_rank == 0) should not start warmup
+    Connecting rank (device_rank == world_size - 1) should not start warmup
     until all other ranks are already WarmingUp.
+    In a 2-node setup, rank 1 is the connecting rank.
     """
     shard0 = get_pipeline_shard_metadata(MODEL_A_ID, device_rank=0, world_size=2)
     shard1 = get_pipeline_shard_metadata(MODEL_A_ID, device_rank=1, world_size=2)
@@ -315,13 +168,13 @@ def test_plan_does_not_start_warmup_for_connecting_rank_until_others_warming():
 
     # Rank 1 is the connecting rank
     bound_instance = BoundInstance(
-        instance=instance, bound_runner_id=RUNNER_1_ID, bound_node_id=NODE_A
+        instance=instance, bound_runner_id=RUNNER_2_ID, bound_node_id=NODE_B
     )
     local_runner = FakeRunnerSupervisor(
         bound_instance=bound_instance, status=RunnerLoaded()
     )
 
-    runners = {RUNNER_1_ID: local_runner}
+    runners = {RUNNER_2_ID: local_runner}
     instances = {INSTANCE_1_ID: instance}
     all_runners = {
         RUNNER_1_ID: RunnerLoaded(),
@@ -329,7 +182,7 @@ def test_plan_does_not_start_warmup_for_connecting_rank_until_others_warming():
     }
 
     result = plan_mod.plan(
-        node_id=NODE_A,
+        node_id=NODE_B,
         runners=runners,  # type: ignore
         download_status={},
         global_download_status={NODE_A: [], NODE_B: []},

src/exo/worker/tests/unittests/test_runner/test_event_ordering.py

@@ -1,208 +0,0 @@
-# Check tasks are complete before runner is ever ready.
-from collections.abc import Iterable
-from typing import Callable
-
-import pytest
-
-import exo.worker.runner.runner as mlx_runner
-from exo.shared.types.api import ChatCompletionMessage
-from exo.shared.types.chunks import TokenChunk
-from exo.shared.types.events import (
-    ChunkGenerated,
-    Event,
-    RunnerStatusUpdated,
-    TaskAcknowledged,
-    TaskStatusUpdated,
-)
-from exo.shared.types.tasks import (
-    ChatCompletion,
-    ChatCompletionTaskParams,
-    ConnectToGroup,
-    LoadModel,
-    Shutdown,
-    StartWarmup,
-    Task,
-    TaskStatus,
-)
-from exo.shared.types.worker.runner_response import GenerationResponse
-from exo.shared.types.worker.runners import (
-    RunnerConnected,
-    RunnerConnecting,
-    RunnerIdle,
-    RunnerLoaded,
-    RunnerLoading,
-    RunnerReady,
-    RunnerRunning,
-    RunnerShutdown,
-    RunnerWarmingUp,
-)
-from exo.utils.channels import mp_channel
-
-from ...constants import (
-    CHAT_COMPLETION_TASK_ID,
-    COMMAND_1_ID,
-    INITIALIZATION_TASK_ID,
-    INSTANCE_1_ID,
-    LOAD_TASK_ID,
-    MODEL_A_ID,
-    NODE_A,
-    RUNNER_1_ID,
-    SHUTDOWN_TASK_ID,
-    WARMUP_TASK_ID,
-)
-from ..conftest import get_bound_mlx_ring_instance
-
-
-def make_nothin[T, U, V](res: T) -> Callable[[], T]:
-    def nothin(*_1: U, **_2: V) -> T:
-        return res
-
-    return nothin
-
-
-nothin = make_nothin(None)
-
-
-INIT_TASK = ConnectToGroup(
-    task_id=INITIALIZATION_TASK_ID,
-    instance_id=INSTANCE_1_ID,
-)
-
-LOAD_TASK = LoadModel(
-    task_id=LOAD_TASK_ID,
-    instance_id=INSTANCE_1_ID,
-)
-
-WARMUP_TASK = StartWarmup(
-    task_id=WARMUP_TASK_ID,
-    instance_id=INSTANCE_1_ID,
-)
-
-SHUTDOWN_TASK = Shutdown(
-    task_id=SHUTDOWN_TASK_ID,
-    instance_id=INSTANCE_1_ID,
-    runner_id=RUNNER_1_ID,
-)
-
-CHAT_PARAMS = ChatCompletionTaskParams(
-    model=str(MODEL_A_ID),
-    messages=[ChatCompletionMessage(role="user", content="hello")],
-    stream=True,
-    max_tokens=4,
-    temperature=0.0,
-)
-
-CHAT_TASK = ChatCompletion(
-    task_id=CHAT_COMPLETION_TASK_ID,
-    command_id=COMMAND_1_ID,
-    task_params=CHAT_PARAMS,
-    instance_id=INSTANCE_1_ID,
-)
-
-
-def assert_events_equal(test_events: Iterable[Event], true_events: Iterable[Event]):
-    for test_event, true_event in zip(test_events, true_events, strict=True):
-        test_event.event_id = true_event.event_id
-        assert test_event == true_event, f"{test_event} != {true_event}"
-
-
-@pytest.fixture
-def patch_out_mlx(monkeypatch: pytest.MonkeyPatch):
-    # initialize_mlx returns a "group" equal to 1
-    monkeypatch.setattr(mlx_runner, "initialize_mlx", make_nothin(1))
-    monkeypatch.setattr(mlx_runner, "load_mlx_items", make_nothin((1, 1, 1)))
-    monkeypatch.setattr(mlx_runner, "warmup_inference", make_nothin(1))
-    monkeypatch.setattr(mlx_runner, "_check_for_debug_prompts", nothin)
-
-    def fake_generate(*_1: object, **_2: object):
-        yield GenerationResponse(token=0, text="hi", finish_reason="stop")
-
-    monkeypatch.setattr(mlx_runner, "mlx_generate", fake_generate)
-
-
-def _run(tasks: Iterable[Task]):
-    bound_instance = get_bound_mlx_ring_instance(
-        instance_id=INSTANCE_1_ID,
-        model_id=MODEL_A_ID,
-        runner_id=RUNNER_1_ID,
-        node_id=NODE_A,
-    )
-
-    task_sender, task_receiver = mp_channel[Task]()
-    event_sender, event_receiver = mp_channel[Event]()
-
-    with task_sender, event_receiver:
-        for t in tasks:
-            task_sender.send(t)
-
-        # worst monkeypatch known to man
-        # this is some c++ nonsense
-        event_sender.close = nothin
-        event_sender.join = nothin
-        task_receiver.close = nothin
-        task_receiver.join = nothin
-
-        mlx_runner.main(bound_instance, event_sender, task_receiver)
-
-        return event_receiver.collect()
-
-
-def test_events_processed_in_correct_order(patch_out_mlx: pytest.MonkeyPatch):
-    events = _run([INIT_TASK, LOAD_TASK, WARMUP_TASK, CHAT_TASK, SHUTDOWN_TASK])
-
-    expected_chunk = ChunkGenerated(
-        command_id=COMMAND_1_ID,
-        chunk=TokenChunk(
-            idx=0,
-            model=MODEL_A_ID,
-            text="hi",
-            token_id=0,
-            finish_reason="stop",
-        ),
-    )
-
-    assert_events_equal(
-        events,
-        [
-            RunnerStatusUpdated(runner_id=RUNNER_1_ID, runner_status=RunnerIdle()),
-            TaskStatusUpdated(
-                task_id=INITIALIZATION_TASK_ID, task_status=TaskStatus.Running
-            ),
-            TaskAcknowledged(task_id=INITIALIZATION_TASK_ID),
-            RunnerStatusUpdated(
-                runner_id=RUNNER_1_ID, runner_status=RunnerConnecting()
-            ),
-            TaskStatusUpdated(
-                task_id=INITIALIZATION_TASK_ID, task_status=TaskStatus.Complete
-            ),
-            RunnerStatusUpdated(runner_id=RUNNER_1_ID, runner_status=RunnerConnected()),
-            TaskStatusUpdated(task_id=LOAD_TASK_ID, task_status=TaskStatus.Running),
-            TaskAcknowledged(task_id=LOAD_TASK_ID),
-            RunnerStatusUpdated(runner_id=RUNNER_1_ID, runner_status=RunnerLoading()),
-            TaskStatusUpdated(task_id=LOAD_TASK_ID, task_status=TaskStatus.Complete),
-            RunnerStatusUpdated(runner_id=RUNNER_1_ID, runner_status=RunnerLoaded()),
-            TaskStatusUpdated(task_id=WARMUP_TASK_ID, task_status=TaskStatus.Running),
-            TaskAcknowledged(task_id=WARMUP_TASK_ID),
-            RunnerStatusUpdated(runner_id=RUNNER_1_ID, runner_status=RunnerWarmingUp()),
-            TaskStatusUpdated(task_id=WARMUP_TASK_ID, task_status=TaskStatus.Complete),
-            RunnerStatusUpdated(runner_id=RUNNER_1_ID, runner_status=RunnerReady()),
-            TaskStatusUpdated(
-                task_id=CHAT_COMPLETION_TASK_ID, task_status=TaskStatus.Running
-            ),
-            TaskAcknowledged(task_id=CHAT_COMPLETION_TASK_ID),
-            RunnerStatusUpdated(runner_id=RUNNER_1_ID, runner_status=RunnerRunning()),
-            expected_chunk,
-            TaskStatusUpdated(
-                task_id=CHAT_COMPLETION_TASK_ID, task_status=TaskStatus.Complete
-            ),
-            # CHAT COMPLETION TASK SHOULD COMPLETE BEFORE RUNNER READY
-            RunnerStatusUpdated(runner_id=RUNNER_1_ID, runner_status=RunnerReady()),
-            TaskStatusUpdated(task_id=SHUTDOWN_TASK_ID, task_status=TaskStatus.Running),
-            TaskAcknowledged(task_id=SHUTDOWN_TASK_ID),
-            TaskStatusUpdated(
-                task_id=SHUTDOWN_TASK_ID, task_status=TaskStatus.Complete
-            ),
-            # SPECIAL EXCEPTION FOR RUNNER SHUTDOWN
-            RunnerStatusUpdated(runner_id=RUNNER_1_ID, runner_status=RunnerShutdown()),
-        ],
-    )

src/exo/worker/tests/unittests/test_runner/test_runner_supervisor.py

@@ -1 +0,0 @@
-# TODO:

src/exo/worker/utils/net_profile.py

@@ -15,7 +15,7 @@ async def check_reachability(
 ) -> None:
     """Check if a node is reachable at the given IP and verify its identity."""
 
-    def _fetch_remote_node_id() -> NodeId | None:
+    def _fetch_remote_node_id() -> str | None:
         connection = http.client.HTTPConnection(target_ip, 52415, timeout=1)
         try:
             connection.request("GET", "/node_id")
@@ -29,17 +29,19 @@ async def check_reachability(
             if body.startswith('"') and body.endswith('"') and len(body) >= 2:
                 body = body[1:-1]
 
-            return NodeId(body) or None
+            return body or None
         except OSError:
             return None
         finally:
             connection.close()
 
-    remote_node_id = await to_thread.run_sync(_fetch_remote_node_id)
-    if remote_node_id is None:
+    remote_node_id_raw = await to_thread.run_sync(_fetch_remote_node_id)
+    if remote_node_id_raw is None:
         return
 
+    remote_node_id = NodeId(remote_node_id_raw)
     if remote_node_id == self_node_id:
+        # Connected to ourselves via loopback - skip
         return
 
     if remote_node_id != expected_node_id:

uv.lock

@@ -334,10 +334,8 @@ dependencies = [
     { name = "hypercorn", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "loguru", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "mlx", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
-    { name = "mlx", extra = ["cpu"], marker = "sys_platform == 'linux'" },
     { name = "mlx-lm", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "networkx", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
-    { name = "openai-harmony", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "protobuf", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "psutil", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "pydantic", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
@@ -376,11 +374,9 @@ requires-dist = [
     { name = "huggingface-hub", specifier = ">=0.33.4" },
     { name = "hypercorn", specifier = ">=0.18.0" },
     { name = "loguru", specifier = ">=0.7.3" },
-    { name = "mlx", marker = "sys_platform == 'darwin'", specifier = ">=0.30.1" },
-    { name = "mlx", extras = ["cpu"], marker = "sys_platform == 'linux'", specifier = ">=0.30.1" },
+    { name = "mlx", specifier = ">=0.30.1" },
     { name = "mlx-lm", specifier = ">=0.28.3" },
     { name = "networkx", specifier = ">=3.5" },
-    { name = "openai-harmony", specifier = ">=0.0.8" },
     { name = "protobuf", specifier = ">=6.32.0" },
     { name = "psutil", specifier = ">=7.0.0" },
     { name = "pydantic", specifier = ">=2.11.7" },
@@ -805,20 +801,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/d4/ff/1e1968f107b4221a98dc26832586b1f646b27ddf3e55c95051c09d751f0a/mlx-0.30.1-cp314-cp314-manylinux_2_35_x86_64.whl", hash = "sha256:d18012d5cf0f013bc4a405cfd1e9d2d28e798f4d2dc4f15aa0fbffff73c02ba2", size = 687114, upload-time = "2025-12-18T01:55:56.506Z" },
 ]
 
-[package.optional-dependencies]
-cpu = [
-    { name = "mlx-cpu", marker = "sys_platform == 'linux'" },
-]
-
-[[package]]
-name = "mlx-cpu"
-version = "0.30.1"
-source = { registry = "https://pypi.org/simple" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/64/51/32903727a68a61e972383e28a775c1f5e5f0628552c85cbc6103d68c0dc4/mlx_cpu-0.30.1-py3-none-manylinux_2_35_aarch64.whl", hash = "sha256:3f5dc2e4d0849181f8253508bb6a0854250483fc63d43ac79ec614b19824b172", size = 8992394, upload-time = "2025-12-18T00:16:13.696Z" },
-    { url = "https://files.pythonhosted.org/packages/0c/74/69c21bb907f3c4064881ab0653029c939ae15fc4e63a5301ef8643cb1d68/mlx_cpu-0.30.1-py3-none-manylinux_2_35_x86_64.whl", hash = "sha256:c9ea6992d8c001e1123dfd3b4d4405ff576c787eec52656ad405e3d033a8be60", size = 10553055, upload-time = "2025-12-18T00:16:16.104Z" },
-]
-
 [[package]]
 name = "mlx-lm"
 version = "0.28.3"
@@ -964,27 +946,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/e2/c1/6dba12fdf68b02a21ac411c9df19afa66bed2540f467150ca64d246b463d/numpy-2.3.4-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:e1708fac43ef8b419c975926ce1eaf793b0c13b7356cfab6ab0dc34c0a02ac0f", size = 18652691, upload-time = "2025-10-15T16:17:46.247Z" },
 ]
 
-[[package]]
-name = "openai-harmony"
-version = "0.0.8"
-source = { registry = "https://pypi.org/simple" }
-dependencies = [
-    { name = "pydantic", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
-]
-sdist = { url = "https://files.pythonhosted.org/packages/3e/92/2d038d096f29179c7c9571b431f9e739f87a487121901725e23fe338dd9d/openai_harmony-0.0.8.tar.gz", hash = "sha256:6e43f98e6c242fa2de6f8ea12eab24af63fa2ed3e89c06341fb9d92632c5cbdf", size = 284777, upload-time = "2025-11-05T19:07:06.727Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/45/c6/2502f416d46be3ec08bb66d696cccffb57781a499e3ff2e4d7c174af4e8f/openai_harmony-0.0.8-cp38-abi3-macosx_11_0_arm64.whl", hash = "sha256:029ec25ca74abe48fdb58eb9fdd2a8c1618581fc33ce8e5653f8a1ffbfbd9326", size = 2627806, upload-time = "2025-11-05T19:06:57.063Z" },
-    { url = "https://files.pythonhosted.org/packages/d3/d2/ce6953ca87db9cae3e775024184da7d1c5cb88cead19a2d75b42f00a959c/openai_harmony-0.0.8-cp38-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:e4f709815924ec325b9a890e6ab2bbb0ceec8e319a4e257328eb752cf36b2efc", size = 2948463, upload-time = "2025-11-05T19:06:48.17Z" },
-    { url = "https://files.pythonhosted.org/packages/fa/4c/b553c9651662d6ce102ca7f3629d268b23df1abe5841e24bed81e8a8e949/openai_harmony-0.0.8-cp38-abi3-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:5cfcfd963b50a41fc656c84d3440ca6eecdccd6c552158ce790b8f2e33dfb5a9", size = 2704083, upload-time = "2025-11-05T19:06:50.205Z" },
-    { url = "https://files.pythonhosted.org/packages/9b/af/4eec8f9ab9c27bcdb444460c72cf43011d176fc44c79d6e113094ca1e152/openai_harmony-0.0.8-cp38-abi3-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:0a3a16972aa1cee38ea958470cd04ac9a2d5ac38fdcf77ab686611246220c158", size = 2959765, upload-time = "2025-11-05T19:06:53.62Z" },
-    { url = "https://files.pythonhosted.org/packages/11/3c/33f3374e4624e0e776f6b13b73c45a7ead7f9c4529f8369ed5bfcaa30cac/openai_harmony-0.0.8-cp38-abi3-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:b4d5cfa168e74d08f8ba6d58a7e49bc7daef4d58951ec69b66b0d56f4927a68d", size = 3427031, upload-time = "2025-11-05T19:06:51.829Z" },
-    { url = "https://files.pythonhosted.org/packages/25/3f/1a192b93bb47c6b44cd98ba8cc1d3d2a9308f1bb700c3017e6352da11bda/openai_harmony-0.0.8-cp38-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:c007d277218a50db8839e599ed78e0fffe5130f614c3f6d93ae257f282071a29", size = 2953260, upload-time = "2025-11-05T19:06:55.406Z" },
-    { url = "https://files.pythonhosted.org/packages/5b/f8/93b582cad3531797c3db7c2db5400fd841538ccddfd9f5e3df61be99a630/openai_harmony-0.0.8-cp38-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:8565d4f5a0638da1bffde29832ed63c9e695c558611053add3b2dc0b56c92dbc", size = 3127044, upload-time = "2025-11-05T19:06:59.553Z" },
-    { url = "https://files.pythonhosted.org/packages/1d/10/4327dbf87f75ae813405fd9a9b4a5cde63d506ffed0a096a440a4cabd89c/openai_harmony-0.0.8-cp38-abi3-musllinux_1_2_armv7l.whl", hash = "sha256:cbaa3bda75ef0d8836e1f8cc84af62f971b1d756d740efc95c38c3e04c0bfde2", size = 2932931, upload-time = "2025-11-05T19:07:01.437Z" },
-    { url = "https://files.pythonhosted.org/packages/8a/c8/1774eec4f6f360ef57618fb8f52e3d3af245b2491bd0297513aa09eec04b/openai_harmony-0.0.8-cp38-abi3-musllinux_1_2_i686.whl", hash = "sha256:772922a9bd24e133950fad71eb1550836f415a88e8c77870e12d0c3bd688ddc2", size = 2996140, upload-time = "2025-11-05T19:07:03.438Z" },
-    { url = "https://files.pythonhosted.org/packages/60/c3/3d1e01e2dba517a91760e4a03e4f20ffc75039a6fe584d0e6f9b5c78fd15/openai_harmony-0.0.8-cp38-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:007b0476a1f331f8130783f901f1da6f5a7057af1a4891f1b6a31dec364189b5", size = 3205080, upload-time = "2025-11-05T19:07:05.078Z" },
-]
-
 [[package]]
 name = "packaging"
 version = "25.0"